A Review of Pavement Performance Deterioration Modeling: Influencing Factors and Techniques

Abstract

Accurate modeling of pavement performance is vital to maintaining safe, reliable, and sustainable transportation infrastructure. This review synthesizes current approaches to pavement deterioration modeling, with emphasis on key influencing factors, performance indicators, and methodologies employed within Pavement Management Systems (PMS). Primary deterioration drivers, including traffic loading and environmental stressors, are analyzed for their impact on degradation patterns. Performance indicators such as the Pavement Surface Evaluation and Rating (PASER), Pavement Condition Index (PCI), and International Roughness Index (IRI) are evaluated for their effectiveness in capturing pavement condition and guiding maintenance decisions. Modeling techniques are broadly categorized into deterministic, probabilistic, and intelligent (machine learning–based) frameworks to illustrate the evolution of predictive approaches. Across these approaches, the notion of symmetry can be interpreted as the balance and consistency achieved between model assumptions, input variables, and predicted pavement behavior, while asymmetry represents deviations caused by uncertainty, variability, and nonlinearity inherent in real-world conditions. Recognizing these symmetrical and asymmetrical relationships helps unify different modeling paradigms and provides insight into how each framework handles equilibrium between accuracy, complexity, and interpretability. The review also highlights persistent challenges in data availability, quality, and standardization. Notably, the increasing adoption of machine learning reflects its capacity to handle high-dimensional and spatiotemporal datasets. Recommendations are proposed to improve the robustness, scalability, and transparency of future deterioration models, thereby enhancing their role in data-driven, resilient, and cost-effective pavement management strategies.

1. Introduction

Road infrastructure is a fundamental component of modern transportation networks and plays a crucial role in economic growth and public safety. Well-maintained roads reduce travel time and increase road safety, facilitating the smooth movement of goods, services, and people [1]. A key element of road infrastructure is pavement, which provides a structural and functional surface for vehicular and pedestrian traffic [2]. The condition of pavement directly influences the overall performance of road networks, affecting ride quality and safety. However, over time, pavement structures deteriorate due to traffic loads, environmental conditions, and aging, leading to reduced serviceability and increased maintenance costs [3]. Since pavement serves as the primary load bearing and wear-resistant component of road infrastructure, its deterioration can significantly impact transportation efficiency. Therefore, proper assessment and predictive modeling of pavement deterioration are essential for optimizing maintenance strategies and ensuring the long-term sustainability of road networks.

Building on this need for accurate forecasting, the growing complexity of transportation systems and the increased demand for resilient infrastructure have underscored the importance of proactive pavement management. Agencies worldwide are now shifting from short-term maintenance scheduling to long-term performance forecasting, guided by predictive modeling and data-driven decision-making [4]. This transition has been supported by the availability of large-scale datasets from sources such as the Long-Term Pavement Performance (LTPP) database and regional Pavement Management Systems (PMS), which have provided valuable empirical evidence for model calibration and validation [5]. These databases, when effectively analyzed, allow transportation agencies to anticipate deterioration trends and plan interventions more efficiently.

Pavement deterioration can be categorized into symmetrical and asymmetrical behaviors; it is symmetrical in its uniform design and structural response under ideal conditions and asymmetrical in the uneven degradation introduced by real-world factors such as traffic loading, climatic variations, and construction quality. Understanding this balance is essential for developing predictive models that accurately represent pavement performance and support effective maintenance planning.

Pavement Management Systems (PMS) have long been employed to evaluate pavement condition, plan maintenance, and allocate resources efficiently [6,7]. Recent developments have emphasized predictive modeling as a key component of PMS, shifting agencies from reactive to proactive strategies. By anticipating future conditions based on traffic, climate, and structural factors, predictive models allow timely interventions that prevent minor surface distresses from escalating into costly structural failures [8,9].

Nevertheless, existing PMS frameworks still face several challenges, including limited data consistency across jurisdictions, subjective condition ratings, and a lack of integrated treatment optimization methods [10]. Furthermore, most PMS applications remain largely descriptive, relying on condition surveys and empirical thresholds rather than continuous predictive updates. To overcome these limitations, researchers have explored advanced deterioration models capable of quantifying uncertainty, simulating nonlinear deterioration behavior, and incorporating spatial–temporal variations [11,12,13].

Advances in machine learning (ML) and deep learning (DL) have introduced powerful tools for pavement performance prediction. These techniques can process large and heterogeneous datasets, identify nonlinear deterioration patterns, and improve forecasting accuracy. Researchers have applied a range of models—including artificial neural networks (ANNs), support vector machines (SVMs), random forests, convolutional neural networks (CNNs), and recurrent neural networks (RNNs)—to tasks such as distress identification, crack detection, and condition rating [14,15,16,17]. While these approaches demonstrate clear advantages, differences in data quality, model interpretability, and transferability across regions highlight ongoing challenges in fully integrating intelligent models into PMS.

Additionally, the majority of ML and DL applications have been directed toward distress detection or image-based assessments, while fewer studies have examined their effectiveness in long-term deterioration forecasting or in linking model outputs to maintenance decision-making [1,18,19]. Integrating these intelligent methods with existing probabilistic and deterministic approaches remains an open research need, particularly for translating model predictions into actionable management strategies.

In contrast to earlier surveys that examined pavement deterioration from narrower perspectives (for instance, focusing only on deterministic models or on isolated environmental factors) [4,20,21], this review adopts a holistic approach. To our knowledge, it is the first to integrate the full spectrum of deterioration drivers, performance indicators, and predictive modeling techniques into a single analysis. By uniting these domains, the review reveals how traffic loads, climate effects, and material characteristics (i.e., the causes of pavement distress) interact with measures of pavement condition (such as IRI, PCI, and PASER) across different modeling frameworks—from traditional empirical models to advanced machine learning algorithms. This integrated perspective provides added value over existing literature, which has typically treated these components in isolation.

To address these gaps, this review provides a comprehensive review of pavement performance deterioration modeling by integrating three core dimensions: influencing factors, performance indicators, and modeling techniques. The study emphasizes the evolution of deterioration modeling from traditional deterministic and probabilistic approaches to advanced data-driven and intelligent methods, highlighting their respective advantages, limitations, and areas of overlap. Specifically, this review

• Examines key factors influencing pavement deterioration, including traffic load, material characteristics, and environmental conditions;
• Evaluates major pavement performance indicators used in condition assessment and management systems;
• Compares and analyzes modeling frameworks, from deterministic, probabilistic, and machine learning-based techniques, to identify research trends and methodological gaps.

By consolidating these perspectives, the review not only clarifies the interrelationships among influencing factors and modeling techniques but also identifies future research directions to improve predictive accuracy, enhance interpretability, and support sustainable, data-driven pavement management practices.

The remainder of this paper is organized as follows: Section 2 explains the review methodology. Section 3, Section 4 and Section 5 then synthesize findings on pavement performance indicators, factors influencing deterioration, and modeling techniques, respectively. Section 6 provides a discussion, and Section 7 concludes with future research directions.

2. Review Methodology

This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines [22] to ensure a transparent selection process. The review focused on synthesizing literature related to pavement performance indicators, factors influencing pavement deterioration, and pavement deterioration modeling, three thematic domains that together define the state of practice and research progress in predictive pavement management.

A comprehensive search was conducted across four major databases Scopus, ScienceDirect, ProQuest, and Google Scholar chosen for their extensive coverage of civil-engineering and transportation infrastructure research. Searches were limited to peer-reviewed, full-text articles in English published between 2008 and 2025, while a few foundational pre-2008 papers were retained for context in early modeling developments.

The search strategy combined keywords describing pavement condition, deterioration, and modeling approaches. A representative Boolean expression used across databases was as follows:

(“pavement performance” OR “pavement condition” OR “asphalt pavement” OR “concrete pavement” OR “pavement management system”)

AND (deterioration OR “service life” OR “remaining service life” OR RSL OR “pavement distress” OR roughness OR IRI OR PCI OR PASER OR “skid resistance”)

AND (model* OR prediction OR forecast* OR evaluation OR assessment)

AND (“machine learning” OR “deep learning” OR “Markov chain” OR Bayesian OR deterministic OR probabilistic OR “fuzzy logic”)

AND (traffic OR AADT OR ESAL OR “axle load*” OR climate OR temperature OR precipitation OR flooding OR “freeze–thaw” OR drainage OR subgrade OR “layer thickness” OR “structural number” OR deflection OR FWD OR TSD OR aging OR maintenance OR rehabilitation)

Searches yielded 4955 records distributed across the three major domains: Pavement Performance Indicators (n = 1145), Factors Influencing Pavement Deterioration (n = 1765), and Pavement Deterioration Modeling (n = 2045). After automated deduplication, 2800 unique records remained for screening. During title and abstract screening, 1420 studies were excluded for the following reasons: irrelevant focus (n = 518), published before 2008 (n = 388), or insufficient dataset or methodological depth (n = 330). The remaining 1380 papers underwent full-text assessment for eligibility. At this stage, 1236 articles were excluded because they were out of scope (n = 518), lacked methodological detail (n = 388), or demonstrated limited rigor (n = 330). In addition to the PRISMA screening protocol, the exclusion criteria were designed to reflect a literature quality assessment framework emphasizing methodological rigor, transparency, and data adequacy. Studies were excluded if they lacked sufficient methodological detail, presented limited validation or analytical rigor, relied on incomplete datasets, or demonstrated weak alignment with the review objectives. This framework ensured that only studies with sound methodological credibility and empirical relevance were incorporated into the final synthesis.

Following this multi-stage process, 144 studies were retained for qualitative synthesis, categorized as follows: Pavement Performance Indicators (n = 38), Factors Influencing Pavement Deterioration (n = 42), and Pavement Deterioration Modeling (n = 64). These articles collectively provide the empirical and methodological basis for the analysis presented in subsequent sections. The complete selection workflow is illustrated in Figure 1 (PRISMA flow diagram), which summarizes the identification, screening, and inclusion process adopted in this review. The overall structure of the manuscript is illustrated in Figure 2, which shows how the review progresses from performance indicators and influencing factors to modeling approaches, followed by synthesis, discussion, and future directions.

3. Pavement Performance Indicators

Pavement Performance Indicators are essential for effective pavement management, as it helps identify surface and structural defects, track deterioration trends, and optimize maintenance strategies [23]. Over the years, various indicator systems have been developed to create a standard for pavement condition assessments, ensuring consistency in decision-making for researchers and transportation agencies. These pavement performance indicators typically address four key aspects: ride quality, surface friction, surface distress, and structural capacity. Some of the widely used systems include the Pavement Condition Index (PCI), PASER (Pavement Surface Evaluation and Rating), the International Roughness Index (IRI), Remaining Service Life (RSL), Present Serviceability Rating (PSR), and the Pavement Quality Index (PQI) [24,25,26,27]. These systems provide objective methods to assess pavement performance and support data-driven infrastructure management and other systems. The reviewed studies span a wide geographic range including the United States, Europe, Asia, and the Middle East.

Table 1. Pavement Performance Indicators used in Studies.

Considered Factors	Pavement Performance Indicator	References
Ride Quality	International Roughness Index (IRI), Ride Quality Index, Overall Performance Condition (OPC), Pavement Condition Rating (PCR), Rutting Depth (RD) Rutting Depth Index (RDI), Future Pavement Surface Condition (FPSC). Next-Generation Pavement Performance Measures (NGPPM)	[28,29,30,31,32,33,34,35,36,37]
Friction	Skid number (SN), Skidding Index, International Friction Index, Skidding Resistance Index (SRI)	[38,39,40,41]
Surface Distress	Pavement Condition Index (PCI), PASER, Crack Index (CI), Surface Distress Index, Alligator Deterioration Index (ADI), Distress Rating (DR), Pervious Concrete Distress Index (PCDI), Pavement Distress Condition Index (PDCI), Future Pavement Surface Condition (FPSC), Rutting Depth Index (RDI)	[32,42,43,44,45,46,47,48]
Structural Capacity	Structural Number (SN), Pavement Structural Strength Index (PSSI), Surface Curvature Index (SCI), Structural Capacity Index, Structural Strength Index (StSI), Structural Adequacy Index (SAI)	[3,49,50,51,52,53]

shows an overview of commonly used pavement performance indicators across numerous studies, highlighting the diversity in assessment approaches and the key aspects of pavement performance they capture, such as ride quality, surface condition, and remaining service life. These indicators not only serve as standardized tools for assessing pavement condition but also act as critical inputs and prediction targets in deterioration modeling frameworks. A relationship further explored in subsequent sections of this review.

These performance indicators provide a foundation for evaluating and comparing pavement conditions. In the following subsections, each category of indicator is discussed, including its significance, typical measurement methods, and any limitations or recent advancements. This analysis will also illustrate how these indicators feed into modeling efforts later on (e.g., many deterioration models use IRI or PCI as key outputs or inputs). By establishing a clear understanding of performance metrics, we set the stage for examining the factors that influence these metrics (Section 4) and the modeling techniques that predict their future values (Section 5).

3.1. Ride Quality

Ride quality is a fundamental component of pavement performance evaluation, as it reflects both the comfort and safety experienced by road users and the impact of pavement conditions on vehicle operating costs. Agencies have traditionally measured ride quality using certified inertial profilers, which provide accurate assessments of surface roughness but are often costly and limited in spatial or temporal coverage [54]. Over the years, several indices have been introduced in the literature, including the Ride Quality Index (RQI), Overall Pavement Condition (OPC), and Pavement Condition Rating (PCR), each offering different perspectives on how pavement smoothness relates to user experience. Despite these alternatives, the International Roughness Index (IRI) has emerged as the universally adopted standard. It is the most frequently applied measure within pavement management systems (PMS) and is consistently referenced in the reviewed studies as the primary indicator of ride quality [55,56,57,58].

International Roughness Index (IRI)

The International Roughness Index (IRI) is one of the most widely adopted measures for assessing pavement surface roughness and its influence on ride quality. Developed in the 1980s by the World Bank [59], the index quantifies road roughness by simulating a vehicle’s response as it traverses the pavement profile [60]. Because of its strong correlation with user perception and ride comfort, IRI has become a global standard in pavement management systems (PMS), where it is commonly used to monitor road performance, evaluate serviceability, and guide maintenance decision-making.

The IRI is calculated from the simulated motion of a standardized quarter-car model, which represents the combined dynamic response of a vehicle’s sprung and unsprung masses. The index is expressed as shown in the equation below:

$$IRI ={\displaystyle \frac{1}{L}} {\int }_0^T\left|\dot{z_s} - \dot{z_u}\right|dt.$$

(1)

where $\dot{z_s}$ and $\dot{z_u}$ represent the vertical velocities of the sprung and unsprung masses, respectively, and L is the profile length.

The IRI is expressed in meters per kilometer (m/km), where lower values correspond to smoother pavements and higher values indicate rougher surfaces. According to World Bank and FHWA standards, pavements with IRI ≤ 1.5 m/km are rated good, 1.5–2.7 m/km are fair, and >2.7 m/km are considered poor, requiring maintenance or rehabilitation.

Traditionally, IRI values are obtained using high-precision instruments such as laser profilers and inertial measurement units [61]. While these methods offer accurate results, they are costly and resource-intensive, limiting their feasibility for large-scale monitoring. Recent advancements in mobile sensing technologies have introduced cost-effective alternatives, including smartphone-based applications that utilize accelerometer and GPS data to approximate IRI values from vibration data [62]. Such approaches are particularly valuable in regions with limited access to specialized equipment, enabling broader adoption of network-level condition monitoring.

The practical importance of IRI extends beyond ride quality. Studies have shown that higher roughness levels increase vehicle operating costs, reduce fuel efficiency, and compromise passenger comfort [60]. Furthermore, rough surfaces amplify dynamic vehicle loading, which accelerates pavement fatigue and structural deterioration [62]. For these reasons, transportation agencies often define IRI threshold values to trigger maintenance or rehabilitation interventions, making it a cornerstone metric for optimizing maintenance strategies.

Despite its widespread use, IRI also has limitations. While it effectively captures surface smoothness, it does not account for underlying structural deficiencies or hidden forms of pavement distress. For example, a roadway may exhibit a low IRI, indicating a smooth surface, yet suffer from subsurface cracking or foundation instability that can lead to costly failures. Additionally, inconsistencies in data collection methods—such as differences in wheel-path measurement (single vs. both wheel tracks) or calibration across devices—can produce variability in reported values [63]. Regional differences in threshold definitions further complicate comparisons between agencies. To address these shortcomings, researchers have emphasized combining IRI with other indices, such as the Pavement Condition Index (PCI) or deflection-based parameters, to achieve a more holistic assessment of pavement health [64].

Recent years have seen the growing integration of IRI into predictive modeling frameworks, supported by advances in machine learning (ML) and deep learning (DL). Studies [65,66,67] have demonstrated the effectiveness of artificial neural networks (ANNs), random forest models, support vector machines (SVMs), k-nearest neighbor (KNN), and convolutional neural networks (CNNs) in forecasting IRI based on inputs such as historical performance data, vehicle-mounted sensors, and even satellite imagery. These approaches not only improve prediction accuracy but also enable scalable, automated monitoring across networks. Building on this, Gopisetti et al. [68] incorporated traffic-induced loading, climate variability, and material properties into IRI-based models, underscoring the need to move beyond surface-level measurements. Similarly, Sandra, A. K. et al. [69] developed a regression-based model that integrated common distresses—cracking, potholes, patching, rutting, and raveling—into IRI prediction. Their approach achieved an R² above 0.85, highlighting the strong relationship between distress severity and roughness, and demonstrating the importance of considering both the extent and severity of distresses in forecasting.

IRI still remains a robust and widely accepted indicator of pavement performance due to its simplicity, reliability, and strong correlation with user experience. However, its limitations underscore the need for integrated frameworks that combine IRI with complementary indicators and leverage advanced modeling techniques. Future research should continue to refine IRI prediction by incorporating distress parameters, traffic and environmental influences, and multimodal data sources, thereby strengthening its role in proactive pavement management and improving the accuracy of long-term deterioration forecasting.

3.2. Friction

Friction is a critical category of pavement performance evaluation. It has a direct relationship with roadway safety and operational performance. Inadequate surface friction increases the risk of skidding and vehicle instability, particularly under wetweather conditions, and has been linked to a considerable proportion of roadway crashes. Studies [70,71,72] in the United States have reported that up to 35% of wet weather accidents are associated with insufficient skid resistance, with as many as 70% considered preventable through improved pavement friction management. This underscores the role of friction as a key determinant in maintenance prioritization and safety-driven pavement management strategies. The frictional capacity of pavements is governed by both micro texture and macrotexture [73]. Microtexture refers to the fine roughness of aggregate surfaces that facilitates tire–pavement adhesion, while macrotexture relates to the larger-scale aggregate configuration that enables water drainage and prevents hydroplaning [74]. The progressive degradation of these surface features through aggregate polishing, binder bleeding, or traffic-induced wear directly reduces skid resistance over time.

To quantify these changes, several indices have been developed. The Skid Number (SN) has been the most widely used metric in the United States, forming the basis of many state-level friction monitoring programs [75]. The International Friction Index (IFI) has been promoted as a standardized global measure, enabling comparisons across different measurement devices and regions. Other measures, such as the Skid Resistance Index (SRI), have also been proposed, though their application is less widespread. While these indices translate surface texture and resistance properties into measurable values, relying on a single metric may overlook the complex interplay between microtexture, macrotexture, and other key conditions affecting friction performance.

Friction measurements are also influenced by weather conditions such as temperature, rainfall, and freeze–thaw cycles, which can alter surface moisture and texture characteristics over time. Recent studies and agency guidelines [76,77,78] have shown that temperature and seasonal variations significantly affect skid resistance values; therefore, temperature- or season-specific correction factors are often applied to improve comparability of friction measurements across different climatic conditions

Skid Number (SN)

Skid number (SN) is a critical pavement performance indicator that measures the skid resistance of road surfaces, influencing safety and maintenance decisions. Historically, the Federal Highway Administration (FHWA) has emphasized the importance of skid resistance since the Highway Safety Act of 1966 leading to systematic data collection and analysis to improve road safety [79]. According to ASTM E274 [80], the Skid Number represents the product of the measured friction factor and a scaling constant, as shown in the equation below:

$$SN = 100 \times f,$$

(2)

where f is the friction factor, defined as the ratio of the horizontal frictional force to the vertical load acting on the pavement surface. The scaling constant 100 is applied to express the result as a whole-number index (typically ranging from 0 to 100) for easier interpretation and comparison across different pavement sections.

Recent studies [66,81] have utilized the Long-Term Pavement Performance (LTPP) database [5] to incorporate maintenance and rehabilitation (M&R) history into performance modeling, enhancing predictive accuracy for pavement conditions. In practice, the Skid Number has become the most established measure of pavement friction, with many state highway agencies relying on locked-wheel skid trailers for systematic data collection across their networks [82,83]. Skid number can be evaluated through a range of laboratory and field devices that simulate tire–pavement interaction under different conditions. Laboratory-based devices such as the British Pendulum Tester (ASTM E303) [84] and the Dynamic Friction Tester (DFT) measure [85].

These surveys generate standardized values that form the basis for identifying sections with deficient skid resistance and prioritizing them for maintenance or rehabilitation. Agencies typically establish investigatory and intervention thresholds, where pavements falling below specific SN levels are flagged for closer inspection or immediate corrective action [86].

The rating system for skid resistance generally classifies pavements with SN ≥ 65 as providing excellent friction and safe driving conditions, values between 45 and 65 as fair, and SN < 45 as poor or potentially hazardous—requiring surface treatment or resurfacing [87,88]. While this practice has proven effective for large-scale safety management, it also underscores certain limitations of skid number assessments, particularly their reliance on device-specific measurements and the temporal gaps inherent in periodic survey programs.

Researchers [85,89] have consistently reported that skid resistance deteriorates over time as a result of traffic loading, aggregate polishing, and environmental exposure. Analyses of LTPP data indicate that asphalt pavements generally exhibit more rapid friction loss than Portland cement concrete surfaces, with deterioration rates further accelerated by high temperatures, precipitation, and freeze–thaw cycles [90]. Seasonal variations are also evident, with lower skid numbers observed during warmer months, emphasizing the need to account for temperature effects when comparing results across different testing periods or regions. Laboratory studies support these findings, showing that the hysteresis component of friction decreases with increasing temperature, while adhesion is reduced on polished surfaces, ultimately leading to a decline in overall skid resistance [85].

However, the Skid Number as an indicator also presents certain limitations. It primarily reflects surface friction under specific test conditions and does not account for subsurface structural integrity, drainage efficiency, or ride quality. Moreover, SN values can vary with testing device type, speed, tire characteristics, and ambient temperature, introducing potential inconsistencies across survey campaigns and agencies [66,87].

Despite these challenges, the skid number remains valuable because of its standardization and its strong linkage to roadway safety outcomes. Its integration into predictive models has enhanced the ability of agencies to anticipate when and where skid resistance will decline below acceptable levels, thereby improving the efficiency of resource allocation. Moreover, combining skid number data with maintenance and rehabilitation histories has been shown to yield more accurate predictions of future pavement conditions, underscoring the importance of longitudinal datasets in supporting evidence-based decision-making. Although some studies [91] argue that skid resistance alone should not determine pavement condition, the Skid Number continues to serve as a cornerstone indicator within comprehensive performance frameworks, ensuring that safety remains central to pavement management practices.

3.3. Surface Distress Indicator

Surface distress represents another key category of pavement performance evaluation, focusing on the identification and quantification of visible deterioration such as cracking, rutting, potholes, bleeding, raveling, and aggregate polishing. These defects are influenced by traffic loading, material properties, and environmental exposure, and they play a critical role in determining the serviceability and remaining life of pavement structures [92]. A variety of indices have been introduced in the literature to capture surface defects, including the Crack Index (CI), Distress Rating (DR), and Pervious Concrete Distress Index (PCDI).

Among the available measures, two indicators have been most widely applied in practice: the Pavement Condition Index (PCI) and the Pavement Surface Evaluation and Rating (PASER) system. The PCI is one of the earliest and most standardized indicators, providing detailed quantification of distress through field surveys, while PASER has been introduced as a more recent alternative that emphasizes simplicity and efficiency [93]. These indices therefore capture both traditional and modern practices, with PCI established as the standard for detailed surveys and PASER widely adopted as a cost-effective alternative for local network evaluations.

3.3.1. Pavement Condition Index (PCI)

The Pavement Condition Index (PCI) is a standardized method for assessing pavement surface conditions based on visual inspection of surface distresses [94]. It was developed by the U.S. Army Corps of Engineers and has since been widely adopted in pavement management systems (PMS) to guide maintenance and rehabilitation decisions [95]. The PCI value is reduced by a cumulative deduct value score that considers the kind, quantity, and degree of discomfort, as well as pavement type according to ASTM D6433-18 [96]. The PCI is expressed as shown in the equation below:

$$PCI=100-maxCDV.$$

(3)

where maxCDV is Maximum cumulative defects.

The PCI method involves trained personnel evaluating pavement sections for different types of surface distresses, including cracking, rutting, potholes, and patching [95]. Each distress type is classified based on its severity and extent, which contributes to an overall PCI score ranging from 0 (failed pavement) to 100 (excellent condition) from ASTM D6433-18 [96].

Traditionally, PCI assessments have been conducted manually, requiring inspectors to survey pavement sections and document observed defects. However, manual inspections can be labor-intensive, time-consuming, and subjective due to human judgment [97].

Various studies [98,99,100,101] have been conducted to improve the efficiency and accuracy of PCI by adapting an image-based survey and automated distress detection using machine learning and deep learning. This approach allows for faster and more consistent pavement evaluations while reducing subjectivity. The ability of deep learning models to detect and classify pavement distresses with high precision has made them an effective tool in pavement management systems. Additionally, automated models based on visual data can seamlessly integrate with Geographic Information Systems (GIS) and pavement management software, further enhancing the efficiency of pavement evaluation and maintenance planning. Habib Shahnazari et al. [102] explored the application of machine learning techniques to enhance PCI estimation by developing models based on artificial neural networks (ANN) and genetic programming (GP). This author’s findings highlight the potential of machine learning-driven PCI estimation, particularly for large-scale pavement management systems where traditional methods may be inefficient. M. Jalal et al. [103] further explored the application of artificial neural networks (ANN) for predicting the Pavement Condition Index (PCI) with a focus on optimizing ANN architectures to enhance accuracy. Their study employed an experimental approach to develop an optimized ANN model capable of predicting PCI with significantly improved performance compared to conventional ANN models. By incorporating qualitative variables such as pavement type and year of construction. These findings align with previous research efforts that have sought to integrate machine learning techniques into pavement condition assessment. A study by Karim et al. [104] evaluated the effectiveness of the Pavement Condition Index (PCI) in assessing road surface conditions and guiding maintenance strategies. The study utilized the PAVER system as described in TM 5-623 [105] to systematically inspect and rate pavement distress, ensuring a standardized approach to pavement management. The research focused on a high-traffic corridor, where PCI was used to quantify pavement deterioration and determine necessary maintenance and rehabilitation (M&R) measures. The study highlighted the importance of PCI in prioritizing road repairs, optimizing maintenance budgets, and ensuring long-term pavement performance.

Despite these advancements, PCI remains primarily a surface-level evaluation metric and does not account for structural integrity or subsurface conditions [98], which are critical for long-term pavement performance. A key limitation of PCI is its reliance on visual distress measurements, which may not fully capture ride quality or pavement roughness. Studies [106] have shown a correlation between PCI and other pavement condition indices, such as the International Roughness Index (IRI), which measures pavement smoothness based on vehicle response to surface irregularities. Since PCI does not explicitly assess pavement ride quality, it is often used in conjunction with IRI and other indicators to provide a more comprehensive pavement evaluation. Additionally, while PCI is widely used by transportation and highway agencies, its effectiveness can be influenced by differences in distress rating criteria among evaluators, leading to potential inconsistencies in condition ratings.

3.3.2. PASER (Pavement Surface Evaluation and Rating)

The Pavement Surface Evaluation and Rating (PASER) system is one of the most widely adopted methods for assessing pavement condition within Pavement Management Systems (PMS), particularly across various state transportation agencies in the United States [reference]. Developed by the University of Wisconsin–Madison, PASER offers a cost-effective and practical approach to monitoring pavement performance through visual inspection of surface distress patterns [107]. The system assigns a numerical score ranging from 1 to 10, with higher scores representing better pavement condition. Ratings are based on observable defects such as cracks, raveling, potholes, and rutting, without the need for advanced instrumentation or structural evaluation [108]. As a result, PASER enables rapid, large-scale pavement assessments with minimal technical complexity [109]. Transportation agencies use PASER to prioritize maintenance and rehabilitation activities. Pavements rated between 1 and 2 are typically considered in critical condition and targeted for immediate reconstruction or heavy rehabilitation. In contrast, pavements with moderate scores (e.g., 5–6) may be recommended for preventive maintenance, such as crack sealing or surface treatments, to prolong their service life and avoid costly repairs in the future. This strategic use of PASER facilitates more efficient resource allocation and supports the long-term sustainability of road networks [110].

Several studies have demonstrated the effectiveness of PASER in combination with other performance indices for optimizing pavement management. For instance, Fakhri et al. [111] employed PASER alongside deflection bowl parameters derived from the Falling Weight Deflectometer (FWD) and the International Roughness Index (IRI). Their findings showed that PASER, when integrated with structural indicators, provided a reliable method for evaluating pavement condition, making it an effective tool for identifying rehabilitation needs and informing maintenance strategies. This integration highlights how PASER, when used alongside other metrics, contributes to a more comprehensive approach to infrastructure management. However, PASER’s limitation as a visual inspection tool, which may not capture subsurface issues, remains a concern for long-term predictive modeling. Saha et al. [32] developed a priority-based optimization model for Pavement Management Systems (PMS) using PASER for city roads. This model synchronizes available budgets with pavement conditions, ensuring that, within a limited annual budget, the overall pavement condition can be optimized. This demonstrated PASER’s role in informing pavement management decisions and maximizing resource allocation. While this approach effectively prioritizes maintenance efforts, it would benefit from further consideration of dynamic factors, such as traffic volume changes or weather patterns, which could influence pavement deterioration over time. Including such variables could make the model more adaptable to real-world complexities. Barzegaran et al. [112] used PASER to predict the International Roughness Index (IRI) for asphalt pavements in Kermanshah, Iran, developing IRI prediction models using regression and Artificial Neural Networks (ANN). Their findings demonstrated that PASER, combined with advanced modeling techniques, offers a cost-effective and rapid method for estimating IRI, helping agencies prioritize maintenance at the network level. This approach is particularly useful for agencies without access to expensive profiling equipment. However, the reliance on visual inspection for PASER introduces potential inconsistencies, and further validation in diverse regions is needed to assess the model’s generalizability.

Despite its widespread use, PASER has certain limitations. Because it relies solely on visual inspection, it may not capture underlying structural deficiencies or subsurface damage. Moreover, its qualitative nature introduces the potential for subjective variability between evaluators. These challenges have motivated efforts to supplement PASER with automated data collection methods and integrate it with other performance indicators in comprehensive pavement management frameworks.

3.4. Structural Capacity

Structural capacity is a fundamental dimension of pavement performance, reflecting the ability of pavement layers and the subgrade to sustain repeated traffic loading over their design life. Unlike functional indicators such as roughness or surface distress indices, which primarily reflect user-perceived performance, structural capacity provides a direct measure of the pavement’s load-bearing adequacy. Literature highlights that functional measures alone can be misleading: pavements with poor surface conditions may still possess sufficient strength, while those appearing smooth may be structurally inadequate [113,114,115]. This distinction underscores the need to integrate structural evaluation into pavement management systems to ensure that both surface condition and underlying strength are addressed in maintenance decisions. The most established indicator of structural capacity is the Structural Number (SN), developed under the AASHTO design methodology. Its widespread use has also inspired the development of related indices such as the Structural Adequacy Index (SAI), Structural Strength Index (StSI), and Surface Curvature Index (SCI) which aim to refine structural assessment or adapt it to nondestructive testing methods. Collectively, these measures form the basis for evaluating structural adequacy in modern pavement management, with the SN continuing to serve as the benchmark against which newer approaches are calibrated.

Structural Number (SN)

The Structural Number (SN) is widely recognized as the foundational indicator of pavement structural capacity, linking pavement layer thicknesses, material properties, and drainage conditions into a single value that reflects the load-bearing adequacy of the system [116]. Originating in the American Association of State Highway and Transportation Officials (AASHTO) [117] design methodology, the SN has provided a framework that translates structural requirements into design layer thicknesses, thereby serving as both a design tool and a performance measure. Its continued presence in pavement research and practice underscores its enduring relevance as the baseline against which newer structural indices are often compared.

The historical significance of SN lies in its ability to bridge design and evaluation. While initially developed to guide pavement thickness design, it has since been adapted to evaluate in-service pavements through the effective structural number (SNeff) [118]. Agencies have incorporated SNeff into pavement management information systems (PMIS) as a means of assessing structural adequacy at the network level, ensuring that maintenance and rehabilitation decisions are informed by both surface condition and underlying structural strength [119]. This shift shows an important evolution: the SN is no longer confined to design calculations but is increasingly viewed as a performance indicator critical to lifecycle management.

The accuracy and practicality of Structural Number (SN) estimation have been strengthened by advances in nondestructive testing methods. The Falling Weight Deflectometer (FWD) remains the most widely applied tool as it provides deflection basin data that can be linked to pavement structural capacity. Building on this foundation, several predictive models—such as those introduced under the COST initiative [120]—have been developed to enhance the translation of FWD measurements into reliable SN estimates for both project- and network-level applications. Several researchers including Rohde et al. [121], Kavussi et al. [122], and Kim et al. [123], developed methods for calculating the effective Structural Number (SNeff) from deflection basin data. Subsequent refinements, such as calibrating these models with Long-Term Pavement Performance (LTPP) data, have enhanced their reliability by addressing biases and incorporating temperature effects. More recently, Traffic Speed Deflectometer (TSD) technology has expanded the feasibility of collecting structural data at highway speeds, enabling agencies to evaluate large networks with minimal traffic disruption [51]. Collectively, these advances illustrate a broader trend in the literature: while the Structural Number (SN) remains conceptually straightforward, its estimation now depends on increasingly sophisticated data collection and modeling techniques.

The integration of SN into maintenance decision-making highlights both its utility and its limitations. On one hand, numerous studies have reported strong correlations between SN values and the performance of maintenance treatments, showing that structurally weak pavements tend to deteriorate more rapidly after intervention [51,124]. This underscores the value of SN as a predictor of treatment longevity and as a prioritization tool within pavement management systems. On the other hand, researchers caution that SN alone cannot fully capture the complex interactions among traffic loads, environmental conditions, and material behavior. Overreliance on SN risks oversimplifying decision-making, pointing to the importance of multi-indicator approaches that integrate structural, functional, and environmental dimensions. Literature reviewed in this section indicates that the Structural Number endures as the benchmark for evaluating pavement structural capacity. Its continued use across decades reflects its practicality, while ongoing refinements through nondestructive testing and predictive modeling seek to mitigate its shortcomings. Thus, SN functions not only as a traditional measure but also as a reference point against which newer indices are calibrated, anchoring the field’s shift toward more holistic and data-driven approaches to pavement evaluation.

4. Factors Influencing Pavement Deterioration

Pavement deterioration is a multifaceted process influenced by a combination of traffic loads, material properties, environmental factors, and maintenance practices [3]. The interaction of these elements determines the rate and mode of pavement distress development. For example, heavy traffic might cause fatigue cracking, which can be accelerated if the pavement structure is weak or if moisture has softened the subgrade. Understanding the role and relative importance of these factors is crucial for developing accurate predictive models that can guide maintenance strategies. By identifying and quantifying these influences, we can improve the precision of deterioration models, leading to more effective infrastructure management and informed decision-making. Figure 3 provides a schematic categorization of the major factors influencing pavement deterioration. It illustrates, for instance, how traffic loads and environmental stressors can induce distresses, how material properties evolve with age, and how structural capacity and maintenance can mediate these effects. In this section, we delve into each category of factors, summarizing key findings from the literature on how they impact pavement performance. By examining how traffic, climate, structure, aging, and maintenance contribute to pavement degradation, we establish the physical mechanisms underlying pavement failure. These mechanisms, in turn, shape the inputs and assumptions used in the predictive modeling approaches discussed later (Section 5).

4.1. Environmental Factors

Environmental conditions—climate and weather—have a significant impact on pavement deterioration. Temperature fluctuations, precipitation, freeze–thaw cycles, and extreme weather events can each trigger or accelerate different forms of pavement distress. Many pavement performance models include basic environmental factors (like average temperature or precipitation) in their predictions, but traditionally, extreme events (e.g., floods, hurricanes, or blizzards) have been less studied and often omitted from models [126,127]. In recent years, however, there has been growing recognition that climate change and increasing weather variability can substantially affect pavements, prompting researchers to call for incorporating more climatic considerations into pavement design and maintenance practices [128,129,130,131].

4.1.1. Temperature Effects

Temperature has a twofold effect on pavements: short-term responses to daily or seasonal temperature changes and long-term aging effects due to prolonged exposure to heat or cold. High pavement temperatures (as in summer heatwaves) tend to soften asphalt binders, making asphalt pavements more prone to rutting (permanent deformation) under traffic. Conversely, in cold temperatures, asphalt binders become brittle, and pavements are more susceptible to cracking (especially thermal cracks). Concrete pavements also experience thermal expansion and contraction, which can cause cracking or joint movement when temperature changes are extreme.

With global temperatures on the rise, many studies have examined how increased heat could impact pavement performance. Bernier et al. [132] reported that high temperatures significantly increased rutting in flexible pavements, with severity varying by region. Zhang et al. [133] highlighted that under Representative Concentration Pathway 8.5 (RCP8.5), ignoring temperature rise reduces pavement resilience. Similarly, Qian Zhang et al. [134] noted that temperature-induced rutting is more severe at higher latitudes, while a study in Germany found that pavements in hotter regions experience greater heat-related distress, necessitating temperature-adaptive materials [135].

In addition to rising temperatures, seasonal fluctuations also play a critical role. Freeze–thaw cycles in cold climates cause repeated expansion and contraction of pavement materials, leading to cracking and structural failure [136]. In warmer regions, sustained high temperatures reduce asphalt stiffness, increasing its vulnerability to deformation under traffic loads [137]. Qiao et al. [138] conducted a sensitivity analysis and found that rising annual temperatures and seasonal fluctuations significantly impact pavement performance. Their study identified temperature as the most influential environmental factor in the Mechanistic–Empirical Pavement Design Guide (MEPDG) [139], highlighting the importance of considering climate factors when developing a pavement deterioration model. Lu Gao et al. [140] further found that both average monthly temperature and precipitation had significant correlations with pavement condition. Their study showed that while short-term temperature effects fluctuated, long-term trends consistently degraded pavement performance. These findings highlight the need for deterioration models that incorporate both rising and seasonal temperature variations to ensure predictive accuracy and structural resilience.

In response to temperature concerns, some adaptive strategies are being considered. Use of polymer-modified binders that have better high-temperature performance (higher softening point) can mitigate rutting in hot climates. In cold regions, using softer binders or additives can improve low-temperature cracking resistance. Additionally, design guidelines now often recommend thicker asphalt layers or better insulation (e.g., use of insulating subbase layers) in permafrost or extreme freeze areas to reduce freeze–thaw damage. Overall, temperature is a dominant environmental factor, and its influence on pavement deterioration is well documented, making it a critical input for any comprehensive pavement performance model.

4.1.2. Precipitation and Flooding

Water is frequently called the “arch-enemy” of pavements. Flooding is a critical environmental factor that accelerates pavement deterioration, particularly by weakening the subgrade and compromising the overall structural integrity of pavements. Vallès-Vallès et al. [141] demonstrated that flooding causes an immediate decline in pavement condition, followed by rapid long-term deterioration. Notably, their research highlighted that roads in better pre-flood conditions tend to deteriorate more quickly after flooding, significantly reducing their service life. This suggests that the resilience of pavements to extreme weather events is crucial in maintaining long-term performance, and it raises the importance of considering flood history when designing or rehabilitating pavements. Barbi, P. S. R. et al. [142] studied the impact of higher precipitation and changes in soil moisture (matric suction) on flexible pavement life. They found that wetter conditions could increase fatigue cracking and rutting damage, estimating that in some climates the lifespan of pavements might be reduced by up to 14 years due to increased precipitation and more intense storm events. This is a substantial effect and aligns with concerns that climate change bringing not just higher temperatures but also more extreme rainfall in many areas could dramatically change infrastructure performance expectations. Their study highlighted that even moderate increases in average moisture, coupled with occasional saturation, can degrade the subgrade and base support enough to manifest earlier and more severe surface distresses.

Another effect of heavy precipitation is the overwhelming of drainage systems. Urban pavements often suffer during intense rainstorms because drainage inlets might not handle the flow, leading to water pooling on the surface. This can cause hydroplaning risks while the water is present, and if water seeps through joints or cracks, it can erode the underlying layers or subgrade (a process known as subgrade softening or even forming voids under the pavement). Similarly, Darshan et al. [143] found that intensified rainfall softens subgrades, thereby reducing their load-bearing capacity and accelerating cracking and rutting. This highlights the need for improved subgrade stabilization methods in areas prone to high rainfall. However, what is less discussed in this study is how localized flooding, often in urban areas with poor drainage systems, can exacerbate these issues. Future research could benefit from examining the interaction between pavement surface condition and localized flooding in densely populated regions. Beyond direct flooding, the intensity of rainfall also overwhelms pavement drainage systems, leading to surface water accumulation, hydroplaning risks, and accelerated surface wear. Ashith et al. [144] emphasized that climate variables such as humidity and precipitation significantly influence permanent deformation and bottom-up cracking, highlighting how variations in climate data affect pavement distress predictions. While this observation is insightful, the long-term effects of climate change on drainage infrastructure require further exploration. Specifically, more data is needed on how increased rainfall frequency impacts existing drainage systems and how these systems can be upgraded to prevent flooding-related pavement degradation.

These studies [141,142,143,144] highlights the multifaceted effects of flooding and precipitation on pavement deterioration, from direct structural damage to the exacerbation of underlying issues like subgrade instability. However, most studies focus on isolated factors such as flooding or rainfall, leaving a gap in understanding how these environmental variables interact over time. Future research could benefit from developing comprehensive models that consider both short-term and long-term climatic influences on pavement degradation, particularly in flood-prone regions.

In summary, precipitation and flooding are formidable contributors to pavement deterioration. They primarily act by weakening the pavement’s supporting layers and by causing physical damage through erosion and freeze–thaw. Their effects can be immediate (after a flood) and long-term (in chronically wet environments). Modern pavement management and design are increasingly accounting for these factors, but this remains an area where the unexpected can happen (a single 100-year flood event can destroy roads that would have otherwise lasted many more years). The literature clearly indicates that neglecting moisture impacts can lead to overly optimistic performance predictions, which is why incorporating these environmental factors into deterioration models (and being prepared with maintenance strategies like better drainage and rapid repair after storms) is so important for infrastructure resilience.

4.2. Structure Factors

Pavement structure is fundamental to how pavement resists traffic loads, environmental stress, and aging. Structural deficiencies often accelerate deterioration, especially when other threats (e.g., moisture, climate variation) interplay with weak pavement structure. The structural composition of pavements including layer thickness, material properties, subgrade strength, and drainage systems play an important role in their deterioration. Structural adequacy determines how well traffic loads are distributed and how effectively pavements resist environmental stresses. Failures in these structural elements have repeatedly been observed to accelerate cracking, rutting, and loss of serviceability.

Layer thickness has long been recognized as a primary structural determinant of pavement performance. Pavements with inadequate thickness often exhibit early fatigue cracking and rutting under heavy traffic [117,145]. A study by Llopis-Castelló et al. [3] suggests that thickness alone is not a guarantee of durability, as even pavements with sufficient structural numbers deteriorate more rapidly under severe climate or traffic variability, underscoring the interaction between structural and environmental factors. In this sense, pavement structure must be understood as part of a system, not a single design parameter.

Material properties further influence how pavement structures respond to stress. Low-quality materials have been linked to severe failures such as potholes and alligator cracking [146], while improved thermal and viscoelastic characteristics help delay deterioration under temperature fluctuations and repeated loading [147]. What emerges from the literature is that structural design is most effective when material selection is tailored to local environmental conditions; for instance, Bilodeau et al. [148] highlight how materials with favorable thermal properties reduce frost damage in cold climates. Thus, the role of materials within the structure is not static but context dependent.

The performance of subgrade soils, which provide foundational support, has also been consistently tied to pavement deterioration. Weak, expansive, or poorly compacted subgrades amplify structural distress, leading to settlement, rutting, and premature cracking [146,148]. While stabilization with additives or geosynthetics has proven effective in improving subgrade strength noted by Amakye et al. [149], the study also shows that poor drainage or saturation can offset these gains. This highlights a recurring theme structural reliability is not simply about initial strength but about maintaining it under changing field conditions.

Drainage integrates directly with structural performance by controlling moisture levels within and beneath the pavement. Poor drainage reduces layer stiffness, weakens subgrades, and accelerates both rutting and stripping [150,151]. Recent studies highlight that innovative drainage solutions such as permeable bases, geotextiles, or permeable pavements can improve pavement resilience [152,153]. However, long-term monitoring shows that the effectiveness of drainage depends as much on maintenance as on initial design, a reminder that structure cannot be separated from broader management practices.

In reviewing structural factors, it becomes evident that no single factor acts in isolation. A pavement is a system, and its deterioration is systemic. Many studies have noted that isolating one variable (e.g., just looking at thickness without considering traffic or environment) gives an incomplete picture [3]. For example, a pavement in a wet climate might need extra structural strength to compensate for the loss of strength during wet periods, something a purely thickness-based comparison might miss. Interdependence is why robust deterioration models and management strategies increasingly use multivariate approaches: including structural capacity (like SN or deflection values), traffic, and environmental parameters together. As highlighted earlier, a recurring theme is that research often addresses one factor at a time (say, effect of subgrade CBR on performance, or effect of drainage on performance), but practical deterioration is caused by combined effects. Hence, the literature suggests that future studies and models should treat these interactions explicitly, such as coupling climate models with structural models or traffic-structure interactions (like how increased truck weights might force a change in required pavement strength).

In conclusion, structural factors (layer thickness, material quality, subgrade support, and drainage) form the backbone of pavement durability. Good structural design and construction can preempt many problems, while deficiencies in structure often manifest later as distresses even if early-life performance was acceptable. When modeling pavement deterioration or planning maintenance, incorporating structural evaluations ensures that we address the root causes of potential failures, not just the symptoms on the surface.

4.3. Pavement Age and Material Properties

As pavements age, their material properties evolve due to environmental exposure, traffic loading, and inherent aging mechanisms, ultimately affecting their structural integrity and service life. Pavement aging leads to physical and chemical changes in asphalt mixtures, resulting in increased stiffness, brittleness, and reduced resistance to cracking. Aging effects are particularly pronounced in the asphalt binder, which undergoes oxidation and volatilization, causing a loss of flexibility [154]. The initial years of service see rapid changes in binder properties, particularly in modified asphalt mixtures, before stabilizing after long-term exposure [155].

The impact of aging is not uniform across pavement layers. While surface layers are more exposed to oxidation, underlying layers also experience mechanical property degradation over time. Studies indicate that after approximately 14 years, mechanical properties, including stiffness and tensile strength, deteriorate consistently across pavement layers, driven by a combination of air void content and binder aging [155]. Additionally, exposure to water exacerbates deterioration by reducing the indirect tensile strength of asphalt, making pavements more susceptible to stripping and fatigue cracking [156].

Aging also alters the dynamic modulus and creep compliance of asphalt mixtures, directly influencing pavement distress. Increased stiffness due to aging correlates with greater rut depth, reduced fatigue resistance, and a higher likelihood of thermal cracking, particularly in colder climates [157,158]. Research has shown that preventive measures, such as applying chip seals and designing thicker pavement layers, can mitigate the effects of aging by enhancing the pavement’s resistance to thermal stress [158].

4.4. Traffic Load and Fatigue in Pavement Deterioration

Traffic loading is one of the most significant factors contributing to pavement deterioration. The repeated stress from vehicle traffic leads to fatigue and eventual structural failure. This process is further influenced by pavement age and environmental conditions, such as precipitation and temperature fluctuations [159]. Notably, heavy vehicle traffic accelerates pavement degradation, with the Equivalent Single Axle Load (ESAL) playing a central role in assessing pavement fatigue and structural distress [160]. ESAL serves as a standardized metric that converts various axle loads into a uniform measure, making it easier to evaluate the cumulative impact of traffic on pavement conditions.

Both static and dynamic loads from repeated vehicle movements contribute to stress variations within the pavement structure, leading to micro-crack initiation and propagation associated with fatigue [161]. This highlights the complexity of traffic-induced damage, which is not just a result of vehicle weight but also the dynamic forces exerted by vehicle movement. The impact of dynamic loads, particularly in high-speed traffic, can exacerbate damage, especially in regions with poor subgrade conditions. This interaction between traffic dynamics and subgrade quality presents a gap in the current literature, where more studies could explore how dynamic forces accelerate deterioration under different conditions.

Several traffic-related parameters are commonly used to assess pavement deterioration. AASHTO [3] defines Annual Average Daily Traffic (AADT) as the total number of vehicles passing a road section per day. While AADT offers an overall measure of traffic volume, it doesn’t distinguish between the damage caused by light vehicles and the more significant effects of heavy trucks. To account for this, the Annual Average Daily Truck Traffic (AADTT) is often used, which quantifies the daily volume of trucks that disproportionately contribute to pavement wear [161]. However, while this measure helps quantify truck traffic, it could be improved by considering not just the volume of truck traffic but also the weight distribution of these trucks across different types and axle configurations. Future research could benefit from developing more granular traffic data that accounts for these variations.

ESAL remains a critical metric in evaluating traffic-induced pavement damage by converting all axle loads into a uniform measure of equivalent load [127]. Studies consistently find that the cumulative effect of high traffic volumes, particularly from heavy trucks, significantly accelerates fatigue cracking, rutting, and other forms of structural failure [162,163,164,165]. This is particularly relevant for pavements that experience high-frequency, heavy truck traffic, which require more frequent maintenance and rehabilitation interventions. However, the direct relationship between traffic loading and pavement deterioration could benefit from further analysis on how traffic patterns, such as seasonal variations or peak traffic periods, affect long-term pavement performance. Exploring the influence of factors like truck type and frequency on pavement wear could lead to more refined models for predicting pavement lifespan and maintenance needs.

While the preceding subsections examined the physical mechanisms through which factors such as environment, structure, material aging, and traffic loading contribute to pavement deterioration, it is equally important to understand how these same variables are operationalized within predictive modeling frameworks. Recent advancements in machine learning and deep learning have enabled the incorporation of diverse pavement-related variables ranging from climatic and structural indicators to maintenance and material characteristics to capture deterioration patterns and forecast long-term pavement performance more accurately.

Table 2. Factors Considered in Models for Pavement Deterioration Prediction.

Studies	Factors Considered	Dataset
Hosseini et al. [166]	Pavement age, reconstruction history, traffic levels, automated distress data, and ride quality metrics	20 years of PMS data (1998–2018, Iowa)
Marcelino et al. [100]	Structural, climatic, and traffic variables with IRI as performance indicator	5–10 years Long-Term Pavement Performance (LTPP) data covering 7 sections
Sidess et al. [43]	PCI-based deterioration influenced by pavement age, load, and condition	Localized PCI records
Gupta et al. [167]	Pavement age, subgrade CBR, thickness, traffic loading, deflection, and roughness	18 low-volume road sections (2 years, India)
Taheri et al. [168]	Structural/material characteristics, compaction density, air voids, traffic loading, precipitation, and freeze–thaw cycles	367 sections from LTPP
Chen et al. [169]	Surface condition, structure, climate, traffic volume, and maintenance actions	LTPP dataset (2464 PCI and 3238 IRI samples)
Deneko et al. [170]	Traffic volume, land use, number of lanes, width, alignment, pavement age, and weather	Pavement survey dataset
Li et al. [171]	Service time, traffic load, rutting, and cracking	Interstate Highway System, Pennsylvania
Ali et al. [172]	Multiple pavement distress types (rutting, fatigue, block, longitudinal, transverse cracking, potholes, patching, delamination)	Field data (2018–2021), 19 roads (37 sections), Newfoundland, Canada
Mahmood et al. [173]	Pavement age, cracking extent, cumulative loading, functional class, climate, and maintenance	In-service pavement test sections, USA
Zhao et al. [174]	Climate and traffic factors for overlay decision-making	LTPP
Al-Samahi et al. [175]	Pavement age, thickness, precipitation, temperature, and Average Annual Daily Truck Traffic (AADTT)	UAE Ministry of Energy dataset + LTPP
Mabureddy et al. [176]	Traffic loading, environmental conditions, and material properties	-

summarizes representative studies, outlining the input factors included, the modeling techniques applied, and the datasets utilized. This overview highlights the variables most frequently used in data-driven models, while also pointing to underexplored factors that remain critical for improving predictive accuracy.

In summary, pavement deterioration emerges from a complex interplay of traffic loads, material properties, environmental stressors, and maintenance practices. Studies consistently show that high traffic volumes and heavy axle loads significantly accelerate fatigue cracking and rutting, especially when the pavement’s structural capacity is marginal. At the same time, climate-related factors like extreme temperature cycles and excessive moisture (e.g., intense rainfall or flooding) weaken pavements and shorten their service life. By considering how these forces act together, researchers gain a holistic understanding of degradation mechanisms. This integrated perspective on deterioration drivers sets the stage for the next section, which examines how pavement condition is measured through performance indicators and rating systems.

5. Approaches to Pavement Deterioration Modeling

Pavement deterioration modeling involves forecasting the progression of surface distresses—such as cracking, rutting, and roughness—over time to support life-cycle cost analysis, maintenance planning, and overall network sustainability. Having established the key performance indicators and influencing factors in previous sections, this section reviews how these elements are integrated into predictive modeling frameworks. Modeling approaches have evolved from early deterministic methods, which use fixed deterioration rates for simplicity but often fail to capture variability, to probabilistic frameworks like Markov chains and survival models that better accommodate uncertainty. More recently, intelligent models—including artificial neural networks, decision trees, and ensemble techniques—have emerged, leveraging large datasets to detect complex, nonlinear deterioration patterns with higher predictive accuracy.

Pavement deterioration models can be classified in various ways. For example, the American Association of State Highway and Transportation Officials (AASHTO) [117] categorizes them into deterministic, probabilistic, Bayesian, and expert-based models, each offering unique benefits depending on data availability and project requirements. Meanwhile, TF Fwa [177] classifies these models into regression analysis models, artificial neural network models, network-level models, project-level models, and probabilistic models. The selection of a modeling approach depends on several factors, including the availability of historical data, pavement type, environmental conditions, and the required level of predictive accuracy [21,164]. For this study, focus is placed on two primary categories. Traditional Modeling Methods rely on historical pavement performance data to predict future conditions but lack the ability to update themselves as new data becomes available. This category includes deterministic and probabilistic approaches. Intelligent Methods continuously refine pavement performance predictions by integrating newly observed data over time, improving accuracy throughout the pavement’s service life.

Regardless of category, all models share a common foundation in the deterioration drivers and performance indicators reviewed earlier

5.1. Traditional Modeling Methods

Traditional methods have been widely employed in pavement deterioration studies due to their interpretability and ease of application. These models have historically underpinned many early Pavement Management Systems (PMS), particularly where structured datasets and analytical simplicity were prioritized. While these approaches offer useful frameworks for assessing long-term pavement performance, they are often limited in their capacity to handle variability and data uncertainty. The section below discusses two common types of traditional modeling methods: deterministic and probabilistic approach.

5.1.1. Deterministic Approach

Deterministic modeling approaches aim to predict future pavement conditions using current and historical data, typically considering the spatial attributes of pavement performance without accounting for uncertainty. These models are crucial for effective pavement management systems (PMS), which help in planning maintenance and optimizing resource allocation. Deterministic models often use mathematical and statistical techniques to predict pavement deterioration, considering factors such as age, traffic load, and environmental conditions. The following sections explore various deterministic approaches and their applications in pavement deterioration modeling.

Deterministic Approaches in Pavement Performance Prediction

Chang et al. [178] emphasized the role of deterministic models in Pavement Management Systems (PMS), particularly in forecasting pavement performance through predefined mathematical functions. These models typically rely on regression-based or mechanistic-empirical relationships to estimate the rate of deterioration. Regression analysis is one of the most widely used techniques in deterministic pavement performance modeling, as it helps establish relationships between pavement condition and influencing factors [179,180]. This approach continuously tracks pavement performance using indicators such as the International Roughness Index (IRI), Pavement Condition Index (PCI), PASER, and Remaining Service Life (RSL). Its popularity stems from its simplicity, practicality, and reliability, as it is grounded in observed pavement behavior. However, since regression models assume that all road segments follow a similar performance trend, variations in pavement conditions across different locations are primarily attributed to the influencing variables incorporated in the model. Fuentes et al. [181] expanded on this by developing deterministic models that utilize objective indicators like IRI and PCI to determine pavement serviceability. Their work highlights how deterministic models aid in maintenance prioritization by offering quantifiable serviceability scores. Similarly, Reza Amin [182] categorized deterministic models into four main types: primary response models, structural models, functional models, and damage models. Each of these focuses on different aspects of pavement performance, ranging from structural integrity to surface condition. While deterministic models provide a structured framework for evaluating pavement deterioration, their accuracy largely depends on the quality of input data and proper model calibration.

In addition to regression analysis, correlation analysis is critical for determining the relationships between pavement performance metrics across time. Khraibani et al. [183] investigated the correlation between pavement performance parameters on the same pavement section using nonlinear mixed-effects models. Their findings demonstrate how structural and environmental variables influence degradation metrics in different pavement sections. Yuan et al. [184] analyzed pavement performance over various segments using a linear mixed-effects model that included time-series and cross-sectional data. This method enhanced the accuracy and resilience of parameter estimates, making it a useful tool for fine-tuning deterministic models in pavement performance prediction.

To better account for the complex nature of pavement deterioration, some studies [185] have employed nonlinear regression models, as pavement condition factors tend to exhibit nonlinear behavior over time [163,173]. Regression analysis in pavement modeling can generally be classified into two main categories based on complexity. The simpler univariate models typically rely on pavement age as the primary predictor of deterioration. In contrast, more advanced multivariate regression models incorporate multiple influencing factors simultaneously, offering a more comprehensive representation of pavement degradation [185,186]. Since pavement deterioration results from the combined effects of various interacting factors, multivariate models provide a more realistic framework for capturing these interactions over time [163,187,188].

Strengths of Deterministic Models

One of the primary advantages of deterministic models is their simplicity and ease of implementation. These models rely on fixed inputs and predefined relationships between variables. The model’s straightforward nature allows for easy interpretation, making it particularly beneficial for agencies managing pavement infrastructure. For instance, the Arizona Department of Transportation (ADOT) utilizes deterministic models, including the marginal cost-effectiveness technique, to guide maintenance and rehabilitation planning with clear, interpretable results [189]. Another key advantage is data efficiency, unlike stochastic models that require extensive datasets to estimate probability distributions. Deterministic models function effectively with limited data. This makes them particularly useful in scenarios where historical pavement condition records are incomplete or unavailable [182]. These models provide practical solutions for pavement management by relying on established relationships between pavement performance indicators and deterioration factors. Furthermore, deterministic models provide predictability and consistency in forecasting pavement deterioration by relying on fixed equations and predefined relationships. This ensures stable and repeatable outcomes, making them ideal for long-term maintenance planning and budgeting. This is used in ADOT’s Pavement Management System (PMS) to support multi-year maintenance strategies by offering clear cost projections and performance expectations [189]. Various studies [30,190,191,192] indicate that deterministic models seamlessly integrate with existing PMS frameworks, enabling both project- and network-level analyses. By providing structured methodologies for assessing pavement conditions and prioritizing maintenance activities based on predefined thresholds, these models enhance decision-making processes.

Limitations of Deterministic Models

Despite their strengths, deterministic models are constrained by several key limitations such as their inability to capture both the uncertainty and nonlinear behavior of pavement deterioration. Pavement performance is influenced by numerous stochastic factors such as variable traffic loading, environmental fluctuations (e.g., freeze–thaw cycles, precipitation), and material inconsistencies, which deterministic formulations treat as fixed. As a result, deterministic models tend to assume simplified relationships, often linear or predefined nonlinear functions that do not reflect the complex, time-dependent interactions among aging, weathering, and repeated loads [171,193]. In fact, several studies note that such models do not consider uncertainty in pavement deterioration associated with traffic and weather conditions [194]. Consequently, deterministic models often yield single point estimates that overlook the range of possible pavement conditions, which may lead to unreliable forecasts and sub-optimal maintenance decisions [195,196].

Another drawback is their limited adaptability. Because deterministic models rely on fixed parameter relationships, they often perform inconsistently when applied across regions with differing climates, traffic regimes, pavement types or material qualities. This rigidity restricts their applicability to large-scale or dynamically evolving networks [197].

Finally, deterministic models do not support risk or reliability assessment, as they produce a single deterministic outcome without quantifying the uncertainty surrounding it. This limitation makes it difficult for decision-makers to evaluate the likelihood of alternative scenarios and to plan for contingencies. In contrast, probabilistic and intelligent modeling techniques explicitly incorporate variability and uncertainty, offering a more comprehensive framework for pavement performance prediction [193,195].

Applications of Deterministic Models in Pavement Management

Despite their limitations, they remain widely used due to their simplicity, consistency, and ease of integration into pavement management systems (PMS). These models help agencies establish baseline deterioration trends and estimate future maintenance needs, making them essential for network-level.

A study [198] was carried out in Tennessee that calibrated treatment performance models using multiple regression analysis, considering variables such as pretreatment Present Serviceability Index (PSI), traffic volume, overlay thickness, and milling depth. This calibration has facilitated the development of more precise performance curves, enhancing the optimization of maintenance planning. Similarly, in South Africa, researchers calibrated the Highway Development and Management (HDM-IV) models using Long-Term Pavement Performance (LTPP) data. Their study demonstrated how local calibration enhances PMS decision-making by aligning model outputs with actual pavement behavior [121]. The Ontario Ministry of Transportation’s PMS was improved by incorporating both deterministic and probabilistic models, optimizing annual investment allocations for maintenance [199].

These studies highlight the effectiveness of deterministic models as fundamental tools in pavement management systems. Their ability to establish baseline deterioration trends and support maintenance planning makes them indispensable for infrastructure management. Furthermore, the integration of deterministic models with advanced techniques, such as probabilistic methods and machine learning, presents a promising avenue for enhancing their predictive accuracy. By refining these models to better capture pavement deterioration complexities, they can evolve into more sophisticated and reliable tools for optimizing long-term pavement performance and investment strategies.

5.1.2. Probabilistic Approach

Probabilistic models differ from deterministic approaches by explicitly incorporating uncertainty and variability into pavement performance prediction. Instead of providing a single forecast, they produce probability distributions or transition likelihoods of future condition states. This makes them particularly suitable in contexts where traffic loading, material variability, and environmental factors are highly uncertain [200]. By quantifying uncertainty, probabilistic approaches enable agencies to evaluate not only the expected condition but also the likelihood of different outcomes, supporting more risk-aware maintenance and rehabilitation planning [201].

Probabilistic Approaches in Pavement Performance Prediction

Several probabilistic frameworks have been applied to pavement maintenance, each grounded in different theoretical foundations.

Markov chain models are the most widely reported in the literature [21]. Wasiq S. et al. [196] emphasized their strength in simulating deterioration as a sequence of condition-state transitions governed by a transition probability matrix (TPM), even when historical data are incomplete. Similarly, Thomas O. et al. [202] highlighted their flexibility and adaptability to different data conditions, which makes them suitable for both project- and network-level analysis. Variants such as homogeneous, staged-homogeneous, non-homogeneous, and semi-Markov chains have been explored by researchers [201,202] to capture different assumptions about time dependence and deterioration progression.

One of the reasons for their popularity is that Markov chains can be directly linked to pavement condition ratings such as IRI, PCI, or PASER, which are often discretized into condition states for modeling. This makes the approach highly compatible with existing pavement management systems (PMS), where condition data are usually stored as categorical or ordinal ratings. Researchers such as Shrestha et al. and Abaza [203,204] noted that this compatibility facilitates integration into maintenance optimization models, allowing agencies to generate transition-based forecasts and directly tie them to treatment policies.

Another advantage discussed in the literature is their computational simplicity. Compared to more complex probabilistic methods, Markov chains require fewer data inputs and can be calibrated relatively easily, even with limited or incomplete datasets. As a result, they are widely adopted in practice by transportation agencies with varying levels of data availability. For example, homogeneous Markov models have been successfully applied where data scarcity made more advanced models impractical, while staged-homogeneous and non-homogeneous variants have been employed when richer datasets were available to reflect time-varying deterioration rates [203,205,206,207].

Recent studies have also extended the framework to semi-Markov models, which relax the assumption that the duration of time in each condition state follows a geometric distribution. As noted by researchers [202], semi-Markov formulations enable more realistic modeling of sojourn times, better capturing the variable pace of deterioration across different pavement types and environments. These refinements illustrate how the Markov framework has evolved to address some of the limitations of the traditional homogeneous formulation while maintaining its practicality.

Despite these developments, Mishalani et al. [208] cautioned that the underlying Markovian assumption that future conditions depend only on the current state and not on past states remains a key limitation. Nevertheless, the balance between simplicity, adaptability, and data efficiency explains why Markov chain models remain the most widely implemented probabilistic technique in pavement deterioration modeling.

Bayesian approaches provide another widely studied framework. Hong et al. [209] demonstrated how Bayesian inference can effectively address heterogeneity in pavement performance datasets, while Mishalani et al. [208] showed how Bayesian models can be integrated with Markov chains to improve predictive reliability. More recent applications, such as those by Biswas et al. [210] and Cui et al. [11] have used Bayesian regression and Bayesian neural networks (BNNs) to forecast roughness and service life while explicitly quantifying uncertainty in the predictions.

Fuzzy regression models have been employed in contexts where imprecise or qualitative data dominate. Wang et al. [211] proposed a hybrid fuzzy–gray model to forecast pavement roughness, while earlier contributions by Zadeh’s fuzzy set theory [212] and applications by subsequent researchers [213,214] showed the usefulness of fuzzy regression in integrating subjective assessments such as surface condition ratings or construction quality.

Sampling and iterative estimation methods have also been applied to reduce bias in parameter estimation. Yehia and Swei [215] emphasized their effectiveness in addressing measurement error and heteroscedasticity in condition data, improving statistical efficiency and reducing unexplained variance in pavement performance models.

Together, these models allow probabilistic modeling to represent both randomness and epistemic uncertainty. As emphasized across these studies their shared advantage lies in producing probability distributions or transition likelihoods, offering more realistic representations of pavement deterioration than deterministic models.

Strengths of Probabilistic Models

The key strength of probabilistic models lies in their ability to quantify uncertainty in deterioration forecasts. Unlike deterministic outputs, which produce fixed trajectories, probabilistic models generate predictive distributions or confidence intervals that support risk-based prioritization and probabilistic performance targets [21,216]. This enables agencies to plan maintenance strategies with an explicit understanding of the likelihood of alternative outcomes. Probabilistic models are also valued for their flexibility in handling incomplete or heterogeneous data. Markov chain models, for instance, have been applied successfully in settings with limited historical data by using TPMs derived from expert judgment or calibration [196,213]. Bayesian approaches extend this flexibility by incorporating expert priorities and continuously updating predictions as new observations become available [217,218]. At the network level, probabilistic formulations have been embedded into optimization frameworks to support budget-constrained scheduling, producing maintenance plans that account for both variability in deterioration and uncertainty in funding [219,220]. Probabilistic models enhance the realism of deterioration modeling by incorporating stochastic representations of damage accumulation, maintenance effectiveness, and environmental effects. For example, staged-homogeneous and non-homogeneous Markov chains better reflect accelerating deterioration in aging pavements [221,222], while Bayesian models can capture regional differences in performance that deterministic models often overlook [223].

Limitations of Probabilistic Models

Despite their advantages, probabilistic models also have several limitations. A widely discussed drawback is the Markovian assumption which states that future pavement conditions depend only on the current state. This assumption can oversimplify real deterioration processes, where cumulative history and past maintenance significantly influence future performance [224]. Similarly, the accuracy of Markov models is sensitive to the number of condition states (NCS) and the length of the duty cycle (LDC); poor choices in discretization can lead to misleading forecasts [221,225].

Another limitation is the high data demand of probabilistic models. Estimating transition probabilities, posterior distributions, or stochastic process parameters requires extensive longitudinal data. In cases where data are sparse, biased, or inconsistent, probabilistic predictions may be unstable or unreliable [221,225]. Calibration of TPMs from deterministic curves is also sensitive to assumptions about variability or scatter, making results inconsistent across studies [70,226]. The complexity of Bayesian framework is a further challenge, as Bayesian neural networks require both advanced statistical expertise and considerable computational resources. These demands can restrict their transferability from research contexts to routine use in agency-level pavement management systems [210]. Furthermore, the interpretability of complex probabilistic models can hinder their practical use. Practitioners accustomed to deterministic indices may find it difficult to apply probabilistic outputs such as posterior distributions or probabilistic decision boundaries without additional explanation [21,227].

Applications of Probabilistic Models in Pavement Management Systems

The integration of probabilistic approaches into pavement management systems (PMS) has expanded significantly, with numerous studies demonstrating their effectiveness in addressing uncertainty in deterioration forecasts and maintenance planning. Rather than providing fixed predictions, these models generate probability-based outcomes that enable agencies to evaluate multiple scenarios and make more informed investment decisions. Applications at the project and network levels illustrate their versatility. Madanat et al. [228] employed probabilistic deterioration models within life-cycle cost analysis, showing that explicitly considering uncertainty in performance thresholds and intervention timing leads to more reliable financial planning. Similarly, Nguyen et al. [229] developed a stochastic optimization framework that allowed network-level maintenance decisions to be prioritized under both condition variability and budgetary constraints, demonstrating how probabilistic forecasts can be embedded within resource allocation strategies.

Researchers have also proposed practical tools to facilitate implementation. Costello et al. [225] developed analytical methods to convert deterministic deterioration curves into transition probability matrices, enabling agencies with legacy deterministic models to adopt probabilistic simulation without overhauling their entire PMS. Yehia et al. [215] applied iterative sampling techniques to Long-Term Pavement Performance (LTPP) data, illustrating how resampling can reduce unexplained variance and enhance the robustness of parametric models used in PMS decision-support.

Beyond methodological refinements, real-world agency use underscores the operational value of probabilistic modeling. The Metropolitan Transportation Commission (MTC) integrated probabilistic performance curves into its PMS, allowing planners to test treatment strategies at different probability thresholds and conduct scenario-based analyses [230]. In Thailand, stochastic deterioration models were applied across the national highway system, improving forecasts for cracking, rutting, and skid resistance and supporting more accurate prioritization of treatments across a diverse pavement network [231].

These applications demonstrate that probabilistic models have matured into essential tools for pavement management, moving beyond academic research to become embedded in practical decision-making frameworks. By incorporating uncertainty into forecasts and linking predictions with financial and operational planning, they enhance transparency and reliability in maintenance decisions. Their integration with optimization techniques and data-driven approaches such as Bayesian inference and Markov-based frameworks further points to a future where maintenance planning is not only uncertainty-aware but also strategically optimized. As emphasized by Yehia et al. [215] and Yamany [216], however, successful deployment requires careful calibration, data quality assurance, and effective communication of probabilistic outputs to ensure that they translate into actionable maintenance strategies.

5.2. Intelligent Models for Pavement Performance Prediction

Intelligent models have been adopted in pavement performance prediction as a response to the limitations of deterministic and probabilistic approaches. Conventional models typically rely on predefined mathematical functions and often impose linear assumptions on deterioration processes. Intelligent approaches include machine learning (ML), deep learning (DL), and hybrid techniques to learn relationships directly from data. This capability allows researchers to account for the combined effects of traffic loading, material properties, environmental conditions, and maintenance history, factors that are difficult to capture adequately using fixed functional forms [169,175].

A consistent finding across recent research is that intelligent models are capable of managing large and heterogeneous datasets, which align well with the demands of modern pavement management systems. Machine learning algorithms including Random Forest, Support Vector Machines (SVM), and Gaussian Process Regression (GPR) have been applied to integrate variables such as pavement age, traffic volumes, climatic conditions, and structural properties, showing improved accuracy in predicting condition ratings [171,173]. DL architectures such as Convolutional Neural Networks (CNNs), used mainly for image-based distress recognition, and Long Short-Term Memory (LSTM) networks, applied to time-series deterioration forecasting, have been effective in capturing spatial and temporal changes in pavement condition [172,174]. In addition, hybrid methods that combine multiple algorithms, for example, ensembles that use genetic algorithms to optimize neural networks or LSTM models enhanced with attention mechanisms, have often reported higher predictive accuracy and reliability than single-model approaches [175,176]. These studies show different intelligent model types offer complementary strengths, with ML providing robust baseline performance, DL capturing more complex data structures, and hybrids seeking to balance both.

The process of developing intelligent models usually follows a structured workflow involving data collection, preprocessing, model training, evaluation, and fine-tuning. Figure 4 illustrates a typical framework, showing how training and testing are incorporated into an iterative cycle before validation and final optimization [232]. Several studies benchmarked intelligent models against traditional regression or probabilistic models and consistently found improvements in predictive accuracy, particularly in network-level applications where conditions vary across regions.

Recent publications emphasize that the contribution of intelligent models lies not only in improving predictive accuracy but also in their capacity to analyze high-dimensional and spatiotemporal datasets. Their adaptability allows them to accommodate variations in data quality and availability, making them suitable for both project- and network-level forecasting. Nonetheless, the literature also points to challenges, particularly on interpretability, the risk of overfitting, and difficulties in transferring models across regions with differing traffic and climate conditions. These limitations highlight the importance of developing explainable AI techniques and embedding intelligent models within broader decision-support frameworks.

Overall, the evidence suggests that intelligent, data-driven models provide a flexible and effective framework for predicting pavement performance. By integrating diverse variables and leveraging advances in computational methods, they not only enhance forecasting accuracy but also support more targeted maintenance planning and resource allocation in pavement management systems.

5.2.1. Machine Learning Model (ML)

Machine learning (ML) approaches have been widely adopted in pavement deterioration modeling because of their ability to capture nonlinear relationships between multiple influencing factors and performance outcomes. Unlike traditional regression-based methods that rely on fixed functional forms, ML models adapt to the data structure and can incorporate diverse variables such as traffic loading, climatic conditions, pavement age, and structural characteristics. These models have been applied to predict key performance indicators including the Pavement Condition Index (PCI), International Roughness Index (IRI), and remaining service life, often with higher accuracy than conventional approaches [66,175].

Scholars have highlighted that different categories of ML algorithms contribute complementary strengths to pavement performance prediction [66,100]. Regression-based models remain useful for their simplicity and interpretability, while ensemble approaches such as Random Forest and Gradient Boosted Trees are valued for their robustness, ability to handle noisy data, and capacity to identify important predictors. Collectively, these methods provide a flexible framework for forecasting deterioration at both the project and network level.

Regression model: Mabureddy et al. [176] developed Random Forest and Neural Network models for predicting crack propagation by integrating traffic load, environmental conditions, and material characteristics. Their models achieved a low RMSE of 1.2 mm/year and an R-squared value close to 0.93, highlighting the capacity of ensemble and neural network techniques to accurately capture complex deterioration patterns over time. Notably, the high performance reported in this study suggests that combining traditional traffic and environmental data with flexible modeling structures yields tangible benefits for predictive accuracy. Al-Samahi et al. [175] proposed a Gaussian Process Regression (GPR) model to forecast the International Roughness Index (IRI) under various climatic conditions. This emphasized the superiority of GPR over traditional models, particularly in its ability to maintain predictive accuracy across diverse environmental contexts.

Ensemble models: Building upon ensemble approaches, Cheng et al. [35] applied Gradient Boosting Decision Trees (GBDT), Random Forest (RF), and Extra Trees algorithms for rutting depth prediction, achieving R-squared values of 0.9761, 0.9833, and 0.9747, respectively. Their study underscores the effectiveness of ensemble methods, which aggregate multiple decision trees to mitigate overfitting and enhance generalization performance, its crucial factor when modeling distress progression over varied pavement networks. Alnaqbi et al. [66] utilized Support Vector Regression (SVR) and Artificial Neural Networks (ANNs) for predicting fatigue cracking in flexible pavements, achieving robust performance metrics with RMSE of 22.416 and R-squared value of 0.80848. Although slightly lower in predictive power compared to GPR or ensemble methods, the application of SVR and ANN still proves the effectiveness of support vector approaches for continuous pavement deterioration trends when carefully optimized for complex, real-world datasets.

Furthermore, Taheri et al. [168] employed ensemble learning techniques including Random Forest, Extremely Randomized Trees (ETR), and XGBoost to predict critical pavement indicators such as wheel path cracking (WpCrAr) and the age of cracking initiation (AgeCrack). By optimizing XGBoost through Bayesian methods, they achieved R-squared values of 0.79 and 0.92, respectively, showcasing the potential of hyperparameter optimization in enhancing model performance for deterioration prediction tasks.

5.2.2. Deep Learning Models (DL)

Turning to deep learning applications, Molinero-Perez et al. [233] developed a comprehensive framework combining Convolutional Neural Networks (CNN) and Feed-forward Neural Networks (FNN) to detect, classify, and predict pavement distresses using vehicle-mounted imagery. Their integration of visual data with traffic and climate information represents a notable advancement, illustrating how multimodal datasets can enhance pavement condition assessment far beyond traditional survey methods.

Similarly, Deng et al. [234] employed Long Short-Term Memory (LSTM) networks to predict pavement performance based on historical maintenance and distress records. Achieving an R-squared value of 0.936, their results highlight LSTM’s superior ability to capture temporal dependencies, making it highly suitable for time-series deterioration modeling—a domain where classical models often fall short.

Emerging research has also explored more advanced architecture. Yao et al. [235] introduced a Transformer-based model incorporating self-attention mechanisms to predict multiple pavement distresses, including rutting and cracking. The model’s ability to dynamically focus on relevant features at each prediction step signifies a major methodological shift, promising higher robustness and interpretability compared to traditional LSTM or CNN-based models. In a related advancement, Philip et al. [236] developed the Attention-Based Selective Embedding Neural Network (ASENN), which demonstrated superior performance across multiple distress datasets. The incorporation of selective embedding layers and attention mechanisms offers a new pathway for efficiently extracting useful features in pavement data, addressing one of the major limitations of earlier deep learning frameworks.

5.2.3. Hybrid Models

Beyond traditional deep learning, hybrid and uncertainty-aware models have also emerged. Cui and Wang [11] utilized Bayesian Neural Networks (BNNs) to model pavement deterioration under climate change uncertainty. By offering predictive intervals rather than point estimates, BNNs enhance risk-based decision-making, an important consideration for long-term infrastructure resilience planning. Xu et al. [237] further improved model optimization through their development of the Adaptive Genetic Algorithm-based Random Forest Neural Network (AGA-RFNN), enhancing predictive accuracy across multiple indicators. Finally, Tao et al. [238] introduced the Differential Evolution Particle Swarm Optimization Back Propagation (DEPSO-BP) neural network for Pavement Condition Index (PCI) prediction, demonstrating strong generalization across highway datasets. These hybrid optimization models indicate a broader trend toward integrating evolutionary computation and deep learning to address the high-dimensionality and complexity of pavement deterioration data.

Table 3. Machine learning techniques used in pavement performance modeling in recent years.

Studies	Country	Model/Method	Modeling Method	Uniqueness
[239]	USA (IOWA)	Deterministic ML	LSTM and Regression model	Used PCI to reflect distress, roughness, and deflection. Trained model on three pavement types for broader applicability.
[100]	-	Probabilistic ML	Random forests algorithm	Developed a generalizable model for network-level PMS. Used both 5- and 10-year datasets for long-term robustness.
[167]	India	Probabilistic ML	Statistical analysis tools and Artificial Neural Network (ANN)	Introduced Maintenance Priority Index (MPI) using CBR and ANN. Aimed at optimizing maintenance scheduling strategies.
[168]	USA (Florida)	Probabilistic ML	Random forest, Extremely Randomized Trees (ETR), and Extreme Gradient Boosting (XGBoost)	Focused solely on Asphaltic concrete. Applied Bayesian-optimized XGBoost for enhanced accuracy.
[170]	China	Probabilistic ML	Fully Connected Neural Network (FCNN), Long Short-Term Memory (LSTM), and LSTM-Attention model	Combined LSTM, FCNN, and Attention models. Used feature extraction on five key performance variables.
[171]	China	Probabilistic ML	Artificial Neural Network (ANN)	Converted distress severity (raveling, rutting, cracks) into PCI. Created a unified distress-based classification framework.
[172]	Canada	Probabilistic ML	Coefficient of determination, Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE)	First to include delamination in pavement condition modeling. Expanded range of distress factors considered in performance prediction.
[240]	USA	Probabilistic ML	Artificial Neural Network (ANN)	Developed a multi-input model handling both numerical and categorical data. Emphasized holistic prediction across diverse input types.
[174]	USA (New Jersey)	Probabilistic ML	Support Vector Regression, Ensemble machine learning methods	Compared multiple ML methods (SVM, RF, GBM, Ensembles). Identified optimal models for rut depth and IRI prediction.

presents a selection of recent studies that utilize various ML algorithms ranging from deep learning models like LSTM to ensemble methods such as Random Forests and XGBoost. Each study is summarized based on its methodological framework and unique contributions, such as the development of maintenance indices, incorporation of long-term datasets, or targeted modeling for specific pavement types.

6. Discussion

This section evaluates the key findings from the reviewed literature. It focuses on the pavement performance indicators commonly used, the influencing variables incorporated into modeling frameworks, and a comparative analysis of traditional and intelligent modeling methods. The gradual progression from deterministic and probabilistic models to more advanced machine learning and deep learning techniques highlights a clear shift toward adaptive data-driven approaches in pavement performance modeling.

Table 4. Key characteristics from traditional modeling and intelligent modeling approach.

Modeling Approach	Category	Key Characteristics
Traditional Modeling	Deterministic Models	Provide reliable results for well-defined pavement segments. Use fixed equations or empirical models. Often oversimplify complex deterioration mechanisms. Lack responsiveness to changes in traffic or climate conditions. Limited adaptability to new data or emerging patterns.
	Probabilistic Models	Represent pavement conditions as a probability distribution across states. Handle uncertainty and variability in performance over time. Require historical data for transition probability estimation. Support maintenance planning under uncertainty.
Intelligent Modeling	AI/ML Based Models	Use machine learning or deep learning to model deterioration behavior. Identify complex, nonlinear relationships between variables. Continuously improving as more data becomes available. Capable of integrating heterogeneous data sources (e.g., images, traffic, weather). Provide early warning for critical pavement failures. Require comprehensive datasets and model training. May require explainability techniques for practical decision-making.

compares the defining features and methodological distinctions of these categories.

In traditional modeling approaches, deterministic models remain well-suited for specific pavement sections where conditions are uniform and well-documented. However, they tend to rely on fixed equations and predefined deterioration curves, which often oversimplify the complex mechanisms influencing pavement performance. By contrast, probabilistic models better capture variability and uncertainty, representing pavement conditions as probability distributions across different states. This makes them more realistic for long-term planning, although they depend heavily on the availability of historical data for calibration.

Intelligent models, encompassing DL- and ML-based approaches, mark a significant leap forward. These models excel at identifying nonlinear relationships among diverse variables and can integrate multiple data sources, such as traffic, climate, and imagery. Intelligent models continuously improve as more data becomes available and can provide early-warning signals for critical pavement failures. However, their implementation requires extensive datasets, careful training, and explainability mechanisms to ensure practical application in real-world pavement management systems.

6.1. Pavement Performance Indicators

Pavement performance indicators form the foundation of deterioration modeling, as they translate complex field conditions into measurable values for analysis. Among the most widely used are the International Roughness Index (IRI), the Pavement Condition Index (PCI), and the PASER score. Each reflects a distinct aspect of pavement health: IRI captures ride quality and user comfort, PCI emphasizes surface and structural distresses based on visual surveys, and PASER provides a simplified rating system that facilitates rapid network-level assessments [24,29,104,239].

While these indicators are commonly used across agencies and studies, a significant portion of the literature treats them in isolation. This fragmented approach can lead to models that overlook important interdependencies among different dimensions of pavement condition. For example, IRI is particularly sensitive to surface roughness caused by traffic-induced deformations but is less effective in detecting fatigue cracking or surface raveling. In contrast, PCI captures a broad range of surface defects through deduct value scoring but relies heavily on resource-intensive visual inspections. PASER offers efficiency and simplicity but introduces subjectivity, with coarser gradation and potential variability between raters. Relying solely on any single metric can obscure how structural, functional, and visual deficiencies co-evolve over time.

Recent research increasingly supports the complementary use of multiple indicators to improve predictive accuracy and model robustness. Studies have reported moderate to strong correlations between IRI and PCI, particularly in contexts where surface distress directly affects ride quality, e.g., Barzegaran et al. [112], Zhang et al. [241], Karim et al. [104] showed that using IRI and PCI together in machine learning models yielded higher R² values and more accurate rehabilitation forecasting than IRI alone. Similarly, Barzegaran et al. [112] found that integrating IRI into PASER-based models improved generalizability across urban and rural segments, highlighting the value of multivariate indicator systems for diverse pavement networks. More advanced frameworks have also emerged. Madeh et al. [242] simultaneously predicted IRI and PCI using machine learning algorithms, finding that multivariate output modeling not only improved prediction accuracy but also revealed that initial IRI values were stronger predictors of long-term performance trends—making IRI a potential early warning signal for network-level planning. Extending this line of work, Wu et al. [243] introduced a deep ensemble learning model optimized with Bayesian methods to jointly predict IRI and the pavement modulus. Their hybrid approach, which combined deep neural networks with decision-tree manifolds (TabNet), demonstrated a prediction accuracy of up to 98.7%, outperforming both standalone deep learning and ensemble models. These findings suggest that joint prediction frameworks are not only feasible but also beneficial for capturing the multifactorial nature of pavement deterioration.

Additionally, Fares et al. [244] developed a composite Pavement Condition Rating (PCR) by integrating three key indicators—IRI, rutting, and cracking—and used machine learning to assess the influence of structural and operational factors on this unified metric. Their results identified annual daily traffic (ADT), base layer thickness, and pavement age as the most influential variables, further reinforcing the value of combining indicators to derive more informative and actionable insights.

Despite these advances, the broader literature still underrepresents the interdependence among pavement performance indicators. Many models continue to treat functional, structural, and visual condition metrics as separate domains, missing opportunities to uncover synergies and feedback loops among them. Bridging this gap through integrated indicator modeling and case-based validation will be essential for advancing pavement management systems toward more holistic, accurate, and adaptive deterioration forecasting.

6.2. Influencing Factors in Pavement Deterioration

As shown in Table 2, pavement deterioration is governed by five core domains: traffic loading, environment/climate, structural capacity and materials, construction quality, and maintenance history. While these factors are well recognized, their representation in predictive models remains inconsistent often explaining more variability in model performance than the algorithms themselves.

Traffic loading is the most consistently modeled variable, typically represented through AADT, AADTT, or ESAL. However, recent studies demonstrate that incorporating axle configuration, truck mix, and dynamic loading significantly improves predictive accuracy compared with static averages.

Environmental and climatic factors such as temperature, precipitation, and freeze–thaw cycles are among the most influential drivers of pavement deterioration, directly affecting material stiffness, moisture susceptibility, and structural integrity. High temperatures soften asphalt layers, while repeated freeze–thaw cycles promote cracking and delamination. Excess moisture or flooding weakens subgrades and accelerates roughness growth by several inches per mile annually.

Structural and material characteristics moderate deterioration under traffic and climatic stress. Pavements with higher structural numbers or stronger subgrades exhibit delayed distress, while thin or poorly bonded layers accelerate cracking. However, many models still condense structure into a single index, overlooking layer interaction, modulus variability, and binder aging.

Construction quality and maintenance history remain the least integrated factors despite their clear impact on performance. Inadequate compaction, poor drainage installation, or deviations in layer thickness often trigger premature failures. Similarly, maintenance data are usually encoded as binary variables rather than incorporating treatment timing, type, and condition at intervention and an oversimplification that limits model realism.

Across the studies shown in Table 2, a recurring limitation is the insufficient treatment of factor interactions. For example, heavy truck traffic amplifies freeze–thaw damage in poorly drained pavements, and maintenance effectiveness depends on the timing and type of intervention. Data inconsistencies, particularly in traffic records and maintenance logs, further constrain the accuracy and generalizability of predictive models.

Taken together, these shortcomings highlight a persistent gap: most deterioration models still represent influencing factors as static and independent, even though pavement performance evolves dynamically through their combined and time-dependent effects. Recent research indicate that incorporating underrepresented variables such as flood exposure, construction quality, and maintenance sequencing can explain an additional 5–10% of the residual variance unaccounted for by traditional models [245].

This review therefore moves beyond cataloging deterioration factors to demonstrate how they should be represented within predictive frameworks. The collective evidence points to the need for dynamic, time-aware representations of pavement variables that capture their interdependencies across traffic, climate, structural, and maintenance domains. Embedding these relationships through sensitivity analysis and data-driven feature attribution enables a more interpretable and mechanistically consistent understanding of pavement behavior over time. By integrating conventional predictors with often-overlooked dimensions—such as construction quality, drainage adequacy, and maintenance history—this review establishes a pathway toward next-generation deterioration models that unite empirical accuracy with engineering insight.

6.3. Comparison of Modeling Approaches

This section evaluates the strengths and limitations of each modeling approach based on reviewed studies, highlighting their performance, suitability, and application in pavement deterioration prediction.

6.3.1. Deterministic Models

Deterministic models remain the most frequently used due to their transparency, ease of calibration, and compatibility with existing Pavement Management Systems (PMS). From the reviewed literature, these models, especially regression-based ones, demonstrate consistent application across various transportation agencies, with notable use of performance indicators such as IRI, PCI, PASER, and RSL. However, from the studies examined, a recurring limitation is their tendency to generalize deterioration behavior using fixed equations, often assuming homogeneity across pavement sections. For instance, although multivariate regression models incorporate several influencing variables, their performance is still heavily dependent on data consistency and local calibration. Studies by Chang et al. [178] and Fuentes et al. [181] illustrate that while deterministic models provide structured deterioration curves, they are largely insensitive to real-time variability in traffic, climate, and construction quality. A key insight from this review is that while deterministic models are reliable for short- to medium-term predictions in well-controlled environments, their accuracy significantly declines in complex or heterogeneous conditions. They also lack uncertainty quantification, making them unsuitable for scenarios that require risk-based maintenance planning.

These models are effective as baseline tools, particularly when historical data is sparse or when interpretability is prioritized over precision. However, their practical use should be complemented by models that can adapt to dynamic pavement behaviors or offer probabilistic insights.

6.3.2. Probabilistic Models

Probabilistic models have proven valuable for capturing the uncertainty and variability inherent in pavement deterioration. Among these, Markov chain models remain the most widely used, primarily due to their simplicity and compatibility with pavement management systems (PMS). They offer a structured way to model deterioration as transitions between condition states using transition probability matrices. However, the assumption of time-invariant transition probabilities remains a limitation, particularly in long-term forecasting where deterioration rates are not constant. Studies that employed non-homogeneous or staged-homogeneous Markov models showed improved realism by allowing transition probabilities to change over time or across specific intervals, but these are rarely implemented in practice due to complexity and data demands.

From a modeling standpoint, the Markov framework offers practical utility for network-level planning, especially when data availability is limited. Still, the model’s inability to estimate the time duration within each state (sojourn time) weakens its predictive capability for maintenance scheduling. Semi-Markov models partially address this by allowing flexible distributions for sojourn time, although they are not yet common in operational settings.

Fuzzy regression models offer a complementary probabilistic perspective by handling imprecise or linguistic databases in regions where expert judgment or subjective evaluations play a key role. These models excel in scenarios with limited quantitative data and are particularly suited for incorporating qualitative inputs like surface condition or construction quality. Hybrid approaches that integrate fuzzy logic with gray theory have demonstrated improved performance in modeling IRI, especially where historical data is sparse. However, fuzzy models are limited by their sensitivity to membership function design and lack of standardization.

Bayesian approaches stand out for their ability to incorporate prior knowledge and update predictions as new data becomes available. This makes them well-suited for adaptive pavement modeling under evolving environmental and traffic conditions. Recent advances in Bayesian neural networks and regionalized Bayesian regression have further improved predictive accuracy and uncertainty quantification. Yet, their high computational demand and interpretability challenges limit their widespread adoption in practice.

Among all probabilistic modeling techniques, Markov chain models offer the most balanced trade-off between interpretability, data efficiency, and predictive performance, particularly at the network level. Their structured transition framework makes them accessible for practical implementation, especially in agencies with limited historical data. In contrast, Bayesian and fuzzy logic approaches introduce greater modeling flexibility and the ability to incorporate expert knowledge and uncertainty; however, they demand higher computational effort and a deeper understanding of probabilistic principles. Moving forward, hybrid frameworks that integrate the transparency of Markov models with the uncertainty quantification capabilities of Bayesian and fuzzy methods present a promising direction for more robust and adaptable pavement deterioration prediction.

6.3.3. Intelligent Models

Recent literature confirms that intelligent models, particularly machine learning (ML) and deep learning (DL) methods, have substantially advanced pavement deterioration prediction by capturing nonlinear and multi-factor interactions that traditional approaches often overlook. Their data-driven nature allows them to adapt to diverse inputs including traffic loading, climate variability, material characteristics, and structural attributes producing higher predictive accuracy at both the project and network levels.

A clear trend in references [244,246,247] is the strong performance of ensemble learning techniques, which consistently outperform single models by combining multiple learners. Studies such as those by Cheng et al. [35] and Mabureddy et al. [176] reported near-perfect predictive accuracies for rutting depth and crack propagation, respectively, underscoring their robustness in heterogeneous conditions. These results suggest that ensembles are particularly effective in handling the irregularities and variability inherent in pavement datasets. However, their growing complexity often introduces trade-offs in interpretability and computational burden, which may constrain real-world adoption.

While support vector machines (SVM) and Gaussian process regression (GPR) remain competitive in smaller or noisier datasets, their application is less common in large-scale deterioration modeling where high-dimensional data dominate. Al-Samahi et al. [175] demonstrated that GPR can outperform neural networks under certain conditions, highlighting that simpler models should not be disregarded in favor of more advanced techniques, particularly when data resources are limited. This underscores the need to match model choice to data context rather than defaulting to complexity.

Deep learning architectures, most notably Long Short-Term Memory (LSTM) networks for temporal data and Convolutional Neural Networks (CNNs) for spatial pattern recognition, have attracted significant attention due to their ability to exploit sequential records and imagery data. Studies such as those by Deng et al. [234] and Molinero-Perez et al. [233] demonstrated how these models can forecast performance trends and extract distress patterns from vehicle-mounted imagery. More recently, Transformer-based models and attention mechanisms have been introduced, which dynamically weight input features to enhance prediction accuracy. These developments indicate a shift toward architectures capable of not only high performance but also improved feature attribution, although their “black box” nature continues to raise questions about interpretability.

Hybrid models represent a promising direction in pavement deterioration prediction, as they combine the strengths of machine learning and deep learning with uncertainty modeling. Approaches such as Bayesian Neural Networks (BNNs) and optimization-enhanced ensembles like AGA-RFNN and DEPSO-BP [237,238] not only deliver high predictive accuracy but also provide uncertainty estimates—an essential feature for long-term infrastructure planning. By embedding risk measures and confidence intervals directly into deterioration forecasts, these models bridge the gap between traditional probabilistic frameworks and modern data-driven techniques, offering decision-makers more reliable tools for prioritizing maintenance and managing resources under uncertainty.

Table 5. Recent studies that utilize various intelligent modeling techniques.

Studies	Model Type	Main Features	Strengths	Weaknesses
[171,174,176,239]	Neural Networks (ANN, LSTM, RNN, DNN, BNN)	Adaptable to large datasets with multiple indicators, learn nonlinear patterns and can process sequential data.	Capture complex interactions between pavement age, traffic, climate, and material variables. Effective for time-series prediction of deterioration rates.	Require large, high-quality datasets; prone to overfitting if data is limited. It is difficult to interpret the influence of individual factors on predictions.
[175,229]	Decision Trees	Splits data into decision nodes for classification or regression tasks.	Easy to implement and interpret; suitable for identifying key variables in small datasets.	Lower predictive accuracy on larger or more complex datasets; sensitive to noisy data.
[11,168]	Ensemble Models (Random Forest, Gradient Boosting, XGBoost)	Combine multiple weak learners to improve overall prediction performance and reduce overfitting.	More robust and generalizable than single decision trees; perform well with heterogeneous pavement and environmental data.	Computationally intensive; model complexity can make it harder to interpret results in practice.
[248,249,250]	Support Vector Machines (SVM, SVR)	Create hyperplanes to separate data classes or predict continuous variables.	Perform well with smaller datasets and can handle high-dimensional variables.	Sensitive to parameter tuning and kernel selection; less effective with large, noisy datasets.
[28,206,236,238]	Hybrid Models (e.g., LSTM-Attention, Bayesian Neural Networks, DEPSO-BP)	Integrate multiple modeling techniques to capture uncertainty, spatial variability, and temporal patterns.	Achieve higher accuracy and robustness by leveraging strengths of each method; capable of quantifying uncertainty for risk-based pavement management.	Require advanced expertise, significant computational resources, and extensive datasets to achieve optimal performance.

summarizes recent studies that utilize various intelligent modeling techniques, including their methodological focus and unique contributions to pavement performance prediction.

From a performance standpoint, intelligent models, especially hybrid and ensemble-based models, outperform traditional models in terms of predictive power and adaptability. However, their practical application is sometimes hindered by high computational demands, the need for extensive training data, and limited interpretability. While deep learning models offer accuracy, simpler ML models such as Random Forests may strike a more practical balance between accuracy and ease of implementation in real-world pavement management systems.

Among the range of intelligent models, hybrid ensemble methods optimized via techniques like Bayesian calibration or genetic algorithms stand out for their ability to handle large datasets while capturing both temporal and spatial patterns. When adapted to local contexts and augmented with uncertainty estimates, these methods define a promising frontier for next-generation pavement deterioration modeling.

6.4. Real-World Adoption and Limitations

Despite the development of advanced deterioration modeling techniques, their adoption in practical pavement management remains relatively limited. As summarized in

Table 6. Implementation of Various Modeling Methods in Engineering Practice.

Year Implemented	Agencies	Models Used
2018	USA: Alaska Department of Transport (DOT) [251]	Regression Analysis
2012	Delaware DOT [30]	Regression Analysis
2009	Florida DOT [252]	Markov Chain Process
1993	Washington DOT [253]	Regression Analysis
2019, 2012	Georgia DOT [254,255]	Survival analysis
2011	Indiana DOT [256]	Regression Analysis
2020	Iowa DOT (At project level) [166]	Long Short-Term Memory (LSTM)
2013	Louisiana Department of Transportation and Development (LADOTD) [185]	Regression Analysis
2011	Nevada DOT [257]	Regression Analysis
1998	Minnesota DOT [258]	Regression Analysis
2025	New Jersey DOT [259]	Regression Analysis
2014	North Carolina DOT [46]	Regression Analysis
2011	North Dakota DOT [260]	Regression Analysis

, many state Departments of Transportation (DOTs) in the United States continue to rely heavily on deterministic approaches, particularly regression-based models, due to their simplicity, transparency, and ease of integration with existing Pavement Management Systems (PMS). Markov chain models are also widely applied, largely because they are embedded in legacy PMS frameworks and support scenario-based decision-making.

In contrast, intelligent data-driven approaches, such as deep learning and hybrid optimization frameworks, have seen only minimal implementation in practice. While numerous theoretical studies demonstrate their superior predictive performance, these models are often perceived as resource-intensive, requiring significant institutional capacity for calibration, interpretation, and long-term data management. Transportation agencies, often constrained by budgetary and staffing limitations, tend to favor models that are “good enough” and easily interpretable, even when technically superior alternatives are available.

Data quality and standardization remain a critical barrier to adoption. Machine learning and deep learning models depend on high-quality, longitudinal, and representative datasets covering traffic, climate, structure, and maintenance histories. Yet, many agencies manage fragmented records, which makes calibration and validation challenging. Poorly standardized datasets also undermine model transferability, leading to weak performance when models are applied across regions with different climates, traffic compositions, or construction practices.

Interpretability and transparency are also persistent challenges. Advanced neural networks and ensemble techniques can operate as “black boxes,” producing outputs that are difficult for engineers, policymakers, and the public to understand. By contrast, regression and decision tree models are more easily explained and justified, which partly explains their continued dominance in practice.

Climate resilience presents an emerging challenge. Current PMS implementations primarily reflect historical climate trends, yet recent climate variability including more frequent freeze–thaw cycles, flooding, and extreme heat has accelerated deterioration beyond expected levels. ML and hybrid models offer opportunities to integrate predictive climate data into pavement design and maintenance planning, but few agencies have yet adopted such approaches.

In summary, while academic literature demonstrates significant progress in predictive accuracy and methodological sophistication, practical adoption remains constrained by issues of data quality, interpretability, and climate variability. Bridging this gap requires developing hybrid frameworks that balance accuracy with transparency, creating data standardization and openness, training multidisciplinary teams, and aligning advanced models with the operational realities of transportation agencies.

7. Conclusions and Future Directions

This review highlighted pavement performance deterioration modeling from multiple perspectives, emphasizing the influence of environmental, traffic, and structural factors, as well as the evolution of modeling techniques. The growing number of road users continues to accelerate pavement degradation, reinforcing the need for accurate, adaptive, and interpretable predictive models within Pavement Management Systems (PMS).

• Influence of Performance Indicators and Data Sources: Most existing models rely heavily on surface condition indicators such as the International Roughness Index (IRI) and Pavement Condition Index (PCI). While these measures are valuable for assessing ride quality and surface distress, they often overlook the structural capacity of pavement layers and safety-related attributes. Moreover, many prediction frameworks have been developed using datasets from the Long-Term Pavement Performance (LTPP) database, which provides extensive historical data for calibration and validation. However, the absence of integrated frameworks that combine surface, structural, and safety performance remains a limitation for transportation agencies seeking holistic and data-driven maintenance strategies.
• Evolution of Modeling Techniques: Modeling has evolved from traditional deterministic and probabilistic frameworks toward intelligent, data-driven approaches. Machine learning (ML) and deep learning (DL) methods have demonstrated superior capability in capturing nonlinear relationships among influencing factors such as traffic loading, pavement age, environmental variations, and treatment history. Despite their accuracy, many intelligent models operate as “black boxes,” offering limited interpretability of variable contributions. This limitation highlights the need for hybrid and explainable models that integrate the transparency of statistical approaches with the predictive power of AI-based techniques.
• Comparative Strengths and Limitations: Each modeling technique exhibits unique strengths and drawbacks. Deterministic models remain straightforward and computationally efficient but lack flexibility in representing stochastic variability. Probabilistic models address uncertainty and reliability assessment but require large, high-quality datasets and significant computational resources. Intelligent models excel at handling multi-dimensional data and complex variable interactions; however, they depend heavily on data availability, model calibration, and standardized validation protocols to ensure generalizability.

Looking ahead, the next generation of pavement performance models should move toward integrated, adaptive, and uncertainty-aware systems that combine data-driven learning with engineering insight. Key research priorities include the following:

• Integration of Multisource Indicators: Develop unified models that combine surface, structural, and safety metrics—such as IRI, rutting depth, skid resistance, and remaining service life—to enable comprehensive performance evaluation.
• Data Quality and Automation: Enhance the completeness and consistency of pavement databases through automated sensing, LiDAR, and real-time monitoring to reduce subjectivity and improve model reliability.
• Climate and Regional Calibration: Incorporate climate-sensitive parameters and localized calibration procedures to improve model adaptability across diverse geographic and environmental contexts.
• Hybrid and Explainable Modeling: Advance interpretable hybrid frameworks that merge the strengths of deterministic, probabilistic, and machine learning techniques while maintaining transparency in decision-making.
• Standardization and Implementation: Establish common performance metrics, validation criteria, and data-sharing protocols across agencies to promote interoperability, benchmarking, and practical adoption.

By advancing these focused research areas, future pavement performance models will evolve into more robust, transparent, and context-aware systems, enabling transportation agencies to plan maintenance activities more proactively, optimize resources, and enhance the resilience and sustainability of roadway infrastructure.