Artificial Intelligence and Digital Twins for Sustainable Waste Management: A Bibliometric and Thematic Review

Abstract

Sustainable waste management is a critical challenge for ecological transitions. Emerging technologies such as Artificial Intelligence (AI) and Digital Twins (DT) offer new opportunities to optimize collection, treatment, and valorization processes, thereby promoting circular economy models. This study adopts an integrated approach to analyze the state of the art and key research trajectories related to the application of these technologies in waste management. Through a bibliometric analysis based on the Scopus database and mapping with VOSviewer (version 1.6.20), three main thematic clusters were identified: (i) predictive and environmental methods, (ii) sustainability and optimization, and (iii) monitoring and environmental impacts. A qualitative analysis of the 20 most-cited articles further revealed six major research areas, including waste forecasting, recycled materials, process digitalization, and intelligent environmental monitoring. The findings indicate a growing convergence among digitalization, automation, and sustainability. The adopted approach enables the mapping of major research directions and emerging interconnections among AI, the circular economy, and predictive management, providing an up-to-date and systemic perspective on the field.

1. Introduction

Waste management represents a complex environmental and infrastructural challenge, directly affecting the quality of life, the health of urban ecosystems, and the sustainability of production and consumption models [1,2]. Global population growth, increasing urbanization, and intensive resource consumption have led to a significant rise in waste generation, with substantial environmental and economic impacts [3,4,5].

In this context, the integration of advanced digital technologies—such as Artificial Intelligence (AI), Machine Learning (ML), and Digital Twins (DT)—is profoundly transforming the waste sector. These technologies offer innovative tools for process optimization, real-time monitoring, and predictive modeling [6,7]. When applied across waste collection, treatment, recycling, and resource recovery (i.e., the process of extracting value from waste materials), they enable more intelligent and adaptive management of material flows, reducing environmental impact and improving service efficiency [8].

Among these, the concept of the Digital Twin, a dynamic virtual replica of a physical asset or process that continuously receives real-time data, is particularly promising and stands to support the operational management of complex systems. By creating a virtual replica of physical processes, it becomes possible to simulate scenarios, optimize decision-making, and adapt treatment and recovery systems based on environmental, logistical, and socio-economic variables [9]. This allows operators to test different management strategies virtually before applying them in the real world.

Integration with AI-based predictive models further enhances the ability to anticipate critical issues, forecast incoming waste volumes, and assess the environmental performance of different management strategies [10].

Consistent with the theoretical framework outlined,

Table 1. Association between constructs and thematic descriptors. Our elaboration.

Construct	Description	References
Global challenge	Waste management as a global issue requiring innovative solutions	[1]
Emerging technologies	AI, machine learning, advanced simulation, and modeling for waste optimization	[3]
Digital Twin	Virtual replicas for simulation, decision support, and system optimization	[5]

summarizes the main conceptual constructs identified in the reviewed literature, linking them to specific thematic descriptors and relevant bibliographic references. This preliminary classification provides an interpretative map of recurring research themes, facilitating the understanding of the research directions that will be further explored in the bibliometric analysis and critical discussion.

The role of AI also emerges across various domains, including the sustainable design of recycled materials, environmental risk assessment, and intelligent environmental monitoring, understood as the use of sensor-based systems, predictive analytics, and geospatial data integration to monitor and respond to environmental conditions in real time [11]. In these areas, the adoption of deep learning models, neural networks, and geospatial analysis supports more informed, data-driven decision-making, thereby strengthening the effectiveness of environmental and urban policies [12].

Building on these premises, the present study aims to provide a systematic and up-to-date analysis of the current state of research on the application of AI and DT in sustainable waste management. The paper is structured to guide the reader through a coherent and methodologically transparent path. The adopted approach consists of a bibliometric analysis of key scientific contributions indexed in the Scopus database [12,13], followed by a thematic evaluation of the 20 most-cited articles [14,15,16,17], with the goal of identifying major conceptual clusters, emerging research lines, and perspectives for technological development.

2. Methodology

The methodology adopted, in synergy with the approach proposed by Campana et al., 2025 [18], is structured into two main phases: systematic literature selection and bibliometric analysis.

A systematic literature selection process was conducted following the PRISMA guidelines, aimed at identifying the most relevant academic contributions to the use of Artificial Intelligence (AI) and Digital Twins (DT) in the context of sustainable waste management. The Scopus database served as the primary source, and a structured set of inclusion and exclusion criteria was applied to ensure thematic coherence and the scientific rigor of the selected publications [19,20].

Scopus was chosen as the sole source for bibliographic data collection [12]. Recognized as one of the most comprehensive global information resources, Scopus covers numerous disciplines and offers high-quality academic content. For this reason, it has progressively established itself as the primary database for conducting bibliometric analyses and systematic literature reviews [13].

In the second phase, a bibliometric analysis was conducted with the support of VOSviewer software, with the aim of mapping the main research trends in the field. This included identifying emerging topics, keyword co-occurrences, and collaboration networks among authors [21].

The integration of the PRISMA-based approach with bibliometric analysis provided a systematic and up-to-date overview of the evolving research landscape on the use of AI and DT in waste management. In particular, a detailed examination of the 20 most-cited articles was conducted to highlight key contributions, prevailing theoretical models, and major future research directions.

2.1. Systematic Literature Selection and Inclusion Criteria

To ensure a systematic and reproducible analysis of the scientific literature on the application of AI and Digital Twins (DT) in sustainable waste management, an approach based on the principles of the PRISMA methodology (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) was adopted [18,19].

The bibliographic search was conducted using the Scopus database, employing an advanced search string constructed as shown in Figure 1:

The search string was constructed to incorporate key synonyms and conceptually related terms frequently used within the field. Boolean operators (OR, AND) were employed to balance breadth and precision, allowing the retrieval of a wide yet thematically consistent set of documents. As such, a single, well-structured query was deemed adequate to encompass the diversity of existing research on the topic.

The initial query, executed without any filters, returned a total of 873 documents. A temporal filter was then applied, restricting the selection to publications from 2015 to 2025 to ensure the currency and relevance of the contributions. This step did not alter the total, which remained at 873 documents.

Next, to ensure the scientific rigor and reliability of the dataset, a document-type filter was applied, limiting the analysis to peer-reviewed journal articles and conference proceedings [22]. This reduced the dataset to 674 documents.

Subsequently, a disciplinary filter was introduced, selecting only publications within the following subject areas: Engineering, Environmental Science, Computer Science, and Materials Science—identified as central to the objectives of this study. After applying this filter, the final dataset consisted of 570 documents.

The selection procedure is summarized in the PRISMA flow diagram shown in Figure 2, which adopts the standard four-phase structure: Identification, Screening, Eligibility, and Inclusion. This framework ensures a transparent and systematic presentation of how the initial dataset was progressively filtered. A simplified overview of the applied criteria is provided in

Table 2. Inclusion and exclusion criteria for document selection. Our elaboration.

Phase	Inclusion Criteria	Exclusion Criteria
Identification	Publications indexed in Scopus database; English language; years 2015–2025	Publications outside time window (pre-2015 or post-2025) (n = 0)
Screening	Peer-reviewed documents: journal articles, reviews, and conference papers	Non-peer-reviewed works: notes, editorials, book chapters, etc. (n = 199)
Eligibility	Belonging to subject areas: Computer Science (COMP), Engineering (ENGI), Environmental Science (ENVI)	Records from unrelated domains (e.g., social sciences, business) (n = 104)
Inclusion	Studies matching all previous filters and containing relevant thematic keywords	—(Final records included: 570)

.

In the identification phase, records were collected through the application of a precisely defined Boolean query. During the screening phase, non-peer-reviewed materials were excluded. The eligibility step entailed a manual assessment of the disciplinary relevance of each document. Finally, in the inclusion phase, only those records that fully met both methodological and thematic requirements were retained for analysis [23,24].

2.2. Bibliometric Analysis

From a bibliometric standpoint, VOSviewer represents a leading tool commonly adopted for the graphical representation and examination of scientific literature networks., which include information such as titles, authors, keywords, and citations [25,26]. Thanks to its versatility, the software enables the identification of research trends, impacts, and evolutionary trajectories through the analysis of citation occurrences and keyword co-occurrences [27].

Within the scientific research landscape, VOSviewer has established itself as an essential tool for visually representing bibliometric data, enabling the recognition of research gaps and thematic concentrations within specific domains and highlighting the most influential sources and authors [28,29]. Operating primarily at an aggregated level, the software is particularly effective in cluster analysis [28].

Specifically, the clustering method available in VOSviewer version 1.6.20 was applied to recurring keywords, with each thematic group assigned a distinctive color [30]. In these visualizations, the size of the nodes (circles) reflects the frequency of keyword usage, while colors represent distinct thematic clusters. The positions along the x and y axes are not semantically significant, which means the visual layout can be rotated or flipped without affecting the interpretation of the underlying relationships [31,32].

The dataset used for this analysis was exported from the Scopus database in .CSV format, ensuring compatibility with the VOSviewer software and allowing for accurate processing of bibliographic metadata. A bibliometric map was then generated based on both author keywords and indexed keywords. To produce a meaningful and readable representation, a filter was applied to include only keywords with a minimum of three occurrences within the selected corpus of articles.

This methodology effectively highlighted the most relevant concepts and areas of the highest semantic density within the research field, providing a clear visualization of emerging terms and thematic clusters present in the scientific literature [33] concerning the use of AI and DT in sustainable waste management.

2.3. Thematic Analysis of the Top-Cited Articles

In addition to the bibliometric mapping, a qualitative thematic analysis was performed on the 20 most-cited articles identified in the corpus. Each article was reviewed in full and thematically coded using an inductive approach, guided by recurring categories such as research objective, AI/ML technique, application domain, and observed or simulated outcomes. This thematic classification was conducted following a qualitative interpretive method inspired by the principles of grounded theory [34], allowing themes to emerge directly from the content of the articles without applying a predefined taxonomy. This process enabled the identification of major research streams, which were subsequently cross-referenced with the bibliometric clusters to ensure consistency, coherence, and conceptual saturation.

3. Results

3.1. Results of Bibliometric Analysis

The bibliometric analysis performed through VOSviewer produced the network visualization presented in Figure 3.

The VOSviewer mapping organizes interconnected keywords into three distinct thematic clusters, each represented by a different color for visual clarity. These clusters represent distinct yet interconnected areas of research in the application of artificial intelligence and digital technologies to sustainable waste management.

The red cluster, located on the left and central part of the map, is strongly focused on data-driven methodologies and environmental variables. This group includes keywords such as machine learning, support vector machine, deep learning, prediction, data interpretation, and artificial neural network, indicating a clear emphasis on computational tools used to analyze and model complex waste-related phenomena. Other keywords like methane, organic matter, and environmental factors suggest a research focus on the chemical and environmental dimensions of waste treatment, particularly in contexts such as landfill emissions and organic waste decomposition.

On the other hand, the green cluster, positioned on the right side of the map, highlights research related to optimization, sustainability, and the circular economy. Key terms such as artificial intelligence, digital twin, circular economy, recycling, zero carbon, and sustainable development goals reflect a growing emphasis on integrating advanced technologies into broader sustainability frameworks. This cluster captures the role of AI and digital simulations in performance assessment, process optimization, and supporting the transition to carbon-neutral waste management strategies.

A third blue cluster, positioned toward the upper part of the map, is more concentrated around human–environment interactions. It includes keywords such as air pollution, human, and environmental monitoring, pointing to research that connects waste management with public health concerns and environmental quality. The presence of water supply within the shared areas of the clusters further suggests an interdisciplinary interest in how waste systems intersect with other resource infrastructures.

Visually, the red cluster appears dense and positioned toward the center-left of the map, indicating a high level of internal connectivity among keywords related to analytical models and algorithmic processing of waste-related data. In contrast, the green cluster is more dispersed and located on the right side of the visualization, illustrating a broader systemic perspective encompassing sustainability goals and strategic innovation. Meanwhile, the blue cluster, though smaller, adds depth by emphasizing social and environmental monitoring dimensions.

Table 3. Rank of co-occurrence of keywords. Our elaboration.

Rank	Keyword	TLS
1	prediction	128
2	artificial intelligence	119
3	artificial neural network	106
4	environmental management	106
5	neural-networks	106
6	machine learning	103
7	recycling	94
8	waste management	90
9	adaptation strategies	69
10	adaptation strategy	68

Note: TLS = Total Link Strength.

summarizes the top 10 most frequently co-occurring keywords identified through this bibliometric analysis, reinforcing the conceptual structure illustrated in the visualization. Leading the ranking is prediction (TLS = 128), a term closely aligned with the red cluster, underscoring the prominence of forecasting techniques in waste estimation and emissions modeling. Keywords such as artificial intelligence (TLS = 119) and artificial neural network (TLS = 106) also reflect the dominance of AI-based methods across the literature. These terms, alongside machine learning and environmental management, form the analytical core of the red cluster’s emphasis on computational modeling.

Notably, recycling (TLS = 94) and waste management (TLS = 90) bridge the red and green clusters, demonstrating the integration of data-driven tools with sustainability-oriented practices. Lower-ranked but conceptually important terms like adaptation strategies and adaptation strategy (TLS = 69 and 68, respectively) suggest a growing interest in flexible, resilience-focused waste management frameworks, potentially intersecting with the blue cluster’s socio-environmental themes.

The interconnections among the clusters suggest that sustainable waste management is increasingly addressed through a multi-layered approach, combining AI-based modeling, real-time environmental monitoring, and strategic planning aligned with sustainability goals. These overlaps reveal the field’s interdisciplinary nature and underscore the relevance of digital innovation in addressing the complexity of circular and low-emission waste systems.

3.2. Scientific Journals and Geographic Distribution of Research Output

Table 4. Top 10 most productive journals ranked by total no. of publications. Our elaboration.

Rank	JT	TNP
1	Waste Management	25
2	Journal of Cleaner Production	25
3	Construction And Building Materials	18
4	Science Of The Total Environment	17
5	Lecture Notes In Networks And Systems	15
6	Sustainability Switzerland	14
7	Journal Of Environmental Management	13
8	Waste Management And Research	11
9	Resources Conservation And Recycling	11
10	Iop Conference Series Earth And Environmental Science	11

Note: JT = Journal title; TNP = Total No. of publications.

displays the top 10 most productive journals in the field of AI-based waste management and environmental applications, ranked by total number of publications. Waste Management and the Journal of Cleaner Production lead the list with 25 publications each, reflecting their central role in disseminating research on predictive modeling, recycling technologies, and smart sustainability practices.

Other notable journals include Construction and Building Materials (18 publications) and Science of the Total Environment (17 publications), which emphasize interdisciplinary approaches to construction waste, environmental quality, and technological innovation. Sustainability (Switzerland) and the Journal of Environmental Management also feature prominently, underlining the growing relevance of AI in both theoretical and applied environmental research.

This ranking highlights the increasing visibility and cross-sectoral interest in machine learning and data-driven solutions for sustainable development. For scholars and practitioners, these journals serve as key reference points for emerging trends, case studies, and methodological advancements in smart waste and resource management.

Table 5. Countries rank with number of documents. Our elaboration.

Country	TNP	TNC	TLS
Australia	39	149	23
Iran	22	139	18
Saudi Arabia	25	167	14
United States	16	52	14
Bangladesh	14	64	12
China	9	17	10
Denmark	8	65	8
Hungary	3	90	7
United Arab Emirates	2	21	6
Netherlands	1	32	4

Note: TNP = Total No. of Publications; TNC = Total No. of Citations; TLS = Total Link Strength.

presents a ranking of countries based on their contributions to the literature on AI and digital technologies in waste management, measured by total number of publications (TNP), total citations (TNC), and total link strength (TLS). The data reveals a geographically diverse research landscape, with a strong presence from both developed and emerging economies.

Australia leads the ranking with 39 publications and a total link strength of 23, indicating not only a high volume of output but also robust international collaboration within the field. This is followed by Saudi Arabia (25 publications, TLS = 14) and Iran (22 publications, TLS = 18), both of which demonstrate active engagement in AI-driven environmental research, with Iran showing particularly high citation impact (139 citations).

The United States ranks fourth in terms of output (16 publications), though with relatively low citation impact (52 publications), suggesting either newer entries into the field or less visibility of published work. Bangladesh (14 publications) and Denmark (8 publications) reflect rising participation from countries with growing interest in sustainability and waste-related technological innovation.

Interestingly, Hungary stands out for its high citation-to-publication ratio (90 citations from only three documents), indicating that its contributions, though limited in quantity, are of significant academic influence. Similarly, countries like the United Arab Emirates and the Netherlands, despite low publication counts, demonstrate meaningful participation, as measured by citation and collaboration metrics (TLS of six and four, respectively).

Overall, the geographical spread highlighted in Table 4 confirms the global relevance of AI in sustainable waste management and the importance of fostering international research networks. The variation in total link strength (TLS) also reflects differing levels of collaborative integration across countries, underscoring opportunities to enhance knowledge exchange and cross-border partnerships in this rapidly evolving field.

3.3. Mapping Key Research Streams Through the Most Cited Contributions

The bibliographic analysis subsequently focused on the 20 most-cited articles [15,16,18] addressing the application of AI and Digital Twins (DT) in sustainable waste management. These publications, presented in

Table 6. State-of-the-art analysis. Our elaboration.

Title	References	No. of Citations	Main Focus	Research Trends
Development of machine learning-based models to forecast solid waste generation in residential areas: A case study from Vietnam	[35]	296	Forecasting urban solid waste generation using ML	Improving predictive accuracy and integrating socio-economic data
Mixture optimization for environmental, economical, and mechanical objectives in silica fume concrete	[36]	283	Optimization of concrete mixtures with silica fume using ML	Multi-objective applications for sustainability and mechanical performance
Modeling the mechanical properties of recycled aggregate concrete using hybrid machine learning algorithms	[37]	269	Mechanical properties of recycled concrete through hybrid ML models	Use of SHAP/PDP for explainability in predictive models
Prediction of carbonation depth for recycled aggregate concrete using ANN hybridized with swarm intelligence algorithms	[38]	257	Carbonation depth in recycled concrete using hybrid ANN models	Robust ML models for durability and environmental degradation
Prediction of thermo-mechanical properties of rubber-modified recycled aggregate concrete	[39]	247	Thermo-mechanical properties of rubber-modified recycled concrete	Predicting performance at high temperatures using ML
Evaluating compressive strength of concrete made with recycled concrete aggregates using machine-learning approach	[40]	228	Estimating compressive strength of recycled concrete	Hybrid ML models for assessing concrete quality
A strength prediction model using artificial intelligence for recycling waste tailings as cemented paste backfill	[41]	199	Strength prediction of mine waste-based backfill	Applying BRT + PSO to reduce experimental testing
Evaluation of mechanical properties of concretes containing coarse recycled concrete aggregates using MARS, M5Tree, and LSSVR	[42]	183	Mechanical properties evaluation with multiple ML models	Comparing AI models for accurate property prediction
Digitalizing the circular economy: Circular economy engineering defined by the metallurgical Internet of Things	[43]	156	Digitalization of the circular economy in metallurgical processes	m-IoT, Digital Twins, and systemic optimization for circular economy
Deep Learning and Artificial Intelligence in Sustainability: A Review of SDGs, Renewable Energy, and Environmental Health	[44]	142	Review on AI and DL use for SDGs, renewable energy, and environmental health	AI for energy optimization, waste management, and environmental health
Forecasting municipal solid waste generation using artificial intelligence modeling approaches	[45]	133	Predicting urban waste generation using ML	ANFIS and kNN optimization for accurate monthly forecasts
Modeling and prediction of regional municipal solid waste generation and diversion in Canada using machine learning approaches	[46]	125	Regional waste generation prediction in Canada	Neural networks for national waste inventories
Analysis of input set characteristics and variances on k-fold cross-validation for a Recurrent Neural Network model on waste disposal rate estimation	[47]	112	Cross-validation effects on RNN models for waste disposal	RNN-LSTM to improve waste disposal prediction accuracy
An insight into machine learning models era in simulating soil, water bodies, and adsorption heavy metals	[48]	109	Review of ML models for heavy metal contamination in soil and water	Developing ML for environmental simulation and contamination management
Potential driving forces and probabilistic health risks of heavy metal accumulation in the soils from an e-waste area	[49]	108	Drivers and health risks from heavy metals in e-waste soils	Random forest for prediction and heavy metal mapping
A hybrid intelligent system for designing optimal proportions of recycled aggregate concrete	[50]	96	Optimizing recycled concrete mix designs using AI	Hybrid models and MOO for enhanced sustainability and strength
Considerations, challenges, and opportunities when developing data-driven models for process manufacturing systems	[51]	92	Data-driven modeling for manufacturing processes	Integrating CRISP-DM and AI for waste valorization
A synthesis framework using machine learning and spatial bivariate analysis to identify drivers and hotspots of heavy metal pollution	[52]	85	Spatial analysis and ML for sources and hotspots of heavy metals	Geospatial approach and RF for agricultural soil management
Predictive modeling of swell-strength of expansive soils using artificial intelligence approaches: ANN, ANFIS, and GEP	[53]	85	AI models for geotechnical properties of expansive soils	GEP and ANN for reliable environmental engineering predictions
A deep learning CNN architecture applied in smart near-infrared analysis of water pollution	[54]	84	CNN for NIR analysis of water pollution in agriculture	Intelligent models for rapid water monitoring

, represent foundational scientific contributions in the field. They provide a comprehensive overview of evolving research trends, influential authors and journals, and thematic priorities addressed in recent years [16,18].

The decision to analyze the most-cited rather than the most recent papers was driven by the intention to avoid imposing two consecutive temporal constraints on the bibliometric analysis—the first stemming from the predefined search period in Scopus and the second which would have resulted from selecting only the most recent publications. Thus, the analysis focuses on contributions that have had the greatest scientific impact, capturing well-established theoretical and methodological foundations in the field [15,17,18].

The analysis of the 20 most-cited articles highlights the emergence of a multidimensional and strongly interdisciplinary research landscape. Intelligent technologies are redefining how waste management challenges are addressed, with applications ranging from waste generation forecasting and recycled material design to environmental simulation and industrial optimization within the circular economy framework.

The first key research area involves the use of AI for waste generation prediction and management. Nguyen et al., 2021 [35] developed machine learning models to accurately estimate urban solid waste generation in residential areas, emphasizing the integration of socio-economic data. Similarly, Abbasi and El Hanandeh, 2016 [45] adopted a combination of Adaptive Neuro-Fuzzy Inference System (ANFIS) and k-Nearest Neighbors (k-NN) to produce reliable monthly forecasts, demonstrating the applicability of such approaches to support urban decision-making. At the regional level, Kannangara et al., 2018 [46] employed multi-layer neural networks to model waste generation and recovery in Canada, while Vu et al., 2022 [47] showcased how Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM), combined with cross-validation techniques, significantly improve waste disposal prediction accuracy.

A second major research area focuses on the use of AI and ML for optimizing recycled materials, particularly in the construction sector. Zhang et al., 2021 [36] and Feng et al., 2022 [39] developed predictive models to determine the mechanical and thermo-mechanical properties of recycled concrete, including rubber-modified variations. Similarly, Tran et al., 2022 [40] and Gholampour et al., 2020 [42] applied advanced techniques such as Multivariate Adaptive Regression Splines (MARS), M5 Model Tree (M5Tree), and Least Squares Support Vector Regression (LSSVR) to evaluate compressive strength. At the same time, Peng and Unluer, 2023 [37] and Liu et al., 2021 [38] used hybrid models based on Artificial Neural Networks (ANN) and nature-inspired optimization algorithms to simulate aspects such as carbonation and environmental durability of concrete, proposing data-driven solutions for sustainable building.

Digitalization for circular economy purposes emerged as another transversal research theme. Reuter, 2016 [43] proposed an integrated model to digitalize metallurgical processes and material life cycles, combining technologies such as DT and metallurgical Internet of Things (m-IoT). Likewise, Zhang et al., 2020 [50] illustrated the effectiveness of multi-objective and hybrid intelligence models in guiding the design of high-performance recycled concrete, contributing to more efficient and sustainable resource management.

The environmental management of electronic waste and heavy metal contamination also emerged as a significant area. Chen et al., 2022 [49] and Yang et al., 2021 [52] applied random forest models and geospatial analysis to identify hotspots and risk sources, while Yaseen, 2021 [48], in a comprehensive review, explored the potential of ML models to simulate contamination in soil and water resources, highlighting new perspectives for environmental monitoring. In the geotechnical domain, Jalal et al., 2021 [53] developed AI models such as ANN, ANFIS, and Gene Expression Programming (GEP) to predict the swell strength of expansive soils, supporting sustainable infrastructure design.

The role of deep learning in environmental monitoring has also been explored in the literature. Fan et al., 2023 [44] analyzed the use of AI to achieve Sustainable Development Goals (SDGs), highlighting applications ranging from waste management to environmental health. Furthermore, Chen et al., 2020 [54] proposed a Convolutional Neural Network (CNN)-based architecture for intelligent detection of water pollution using Near-Infrared (NIR) analysis, offering a promising solution for sustainable agriculture and water quality management.

Finally, growing attention has been paid to the integration of AI, DT, and industrial systems. Fisher et al., 2020 [51] presented a framework that combines data-driven methodologies with the Cross-Industry Standard Process for Data Mining (CRISP-DM) to develop predictive models in production processes, with a particular focus on waste valorization and decision support in industrial contexts.

To complement the thematic categorization, a temporal analysis of the most-cited contributions from 2016 to 2023 was conducted to capture the dynamic evolution of research in this field. This dimension addresses the reviewer’s concern by highlighting how scholarly attention has shifted over time in response to technological advances and policy priorities.

The line graph below (Figure 4) illustrates the emergence and evolution of five major research themes: waste forecasting, recycled materials, environmental monitoring, digitalization and circular economy, and AI and industrial systems. The data reveal three distinct phases: an initial focus on forecasting and isolated applications of AI (2016–2018), followed by an expansion into recycled materials and optimization (2019–2021), and finally, a surge in environmental and industrial integration with Digital Twins and deep learning (2021–2023).

This trend indicates a growing systemic perspective, where AI and DT are not only used for prediction or optimization but are increasingly embedded within environmental governance, industrial valorization processes, and sustainability frameworks.

In support of this visualization,

Table 7. Chronological distribution of top-cited articles by thematic area. Our elaboration.

Theme	Timeline	References
Digitalization and Circular Economy	2016	[43]
Waste Forecasting	2016–2019	[45,46,47,50]
AI and Industrial Systems	2020–2021	[51]
Recycled Materials	2020–2023	[39,41,43,50,55]
Environmental Monitoring	2021–2022	[48,49,52,53]

summarizes the temporal distribution of the most-cited articles, organized by theme. This format emphasizes the continuity and expansion of specific research areas and provides direct references to key contributions over time.

This integrated temporal and thematic analysis contributes to a more nuanced understanding of how research on AI and Digital Twins for sustainable waste management has matured and diversified.

3.4. Summary of Thematic and Conceptual Clusters

Overall, the analyzed literature highlights a growing focus on the application of digital and intelligent solutions to address environmental challenges in waste management. AI is emerging not only as a predictive tool but also as a strategic enabler for the digitalization of material flows, real-time environmental monitoring, optimization of production cycles, and sustainable design. Emerging technological trajectories indicate a shift towards predictive and integrated models capable of reducing environmental impacts, valorizing waste as a resource, and supporting the transition to a circular economy.

This study is characterized by its integrated methodological approach, which combines bibliometric analysis with qualitative thematic review. This synergy enabled a comprehensive understanding of current research trajectories, mapping areas of high scientific intensity and identifying interdisciplinary connections between predictive modeling tools, digital simulation, and sustainability strategies. In particular, the emphasis on the role of AI, DT, and hybrid ML models offers an original contribution aimed at promoting a systemic and data-driven perspective on waste management.

Table 8. Summary of constructs. Our elaboration.

Macro Theme	References	Associated Keywords
AI for forecasting and managing waste generation	[35,45,46,47]	waste management, urban waste, prediction, machine learning
AI and ML for recycled materials and sustainable concrete	[36,37,38,39,40,42]	recycling, artificial intelligence, optimization
Digitalization and circular economy	[43,50]	digital twin, circular economy
E-waste, heavy metals, and environmental management	[48,49,52,53]	air pollution, human, environmental monitoring
Deep learning and environmental monitoring	[44,54]	deep learning, sustainable development goal, zero carbon
Integration of AI, industrial systems, and DT	[51]	digital twin, artificial intelligence

summarizes the results of the analysis in a structured format, associating each identified macro-theme with key authors and relevant keywords selected exclusively from those emerging in the bibliometric map (frequency ≥ 3). This organization highlights the main research lines, the most conceptually dense areas, and the semantic relationships among the thematic clusters. The combined approach, integrating conceptual clusters and keyword co-occurrence analysis, has provided a coherent and structured representation of the current scientific landscape, confirming the multidimensional and interdisciplinary nature of digital applications in sustainable waste management.

To complement the thematic clusters and enrich methodological understanding,

Table 9. Summary of AI/ML methods by research area. Our elaboration.

Research Area	AI/ML Methods and Purpose	References
AI for forecasting and managing waste generation	ANN, ANFIS, k-NN, RNN-LSTM—regression techniques to estimate urban and regional waste generation with high temporal accuracy.	[35,45,46,47]
AI and ML for recycled materials	MARS, LSSVR, M5Tree, ANN + PSO/GA—regression and optimization models for compressive strength, carbonation, and durability prediction.	[37,38,39,40,41,43,50,55]
Environmental monitoring	Random Forest, CNN, geospatial ML—classification and spatial simulation to identify pollution hotspots, heavy metals, and water contamination.	[48,49,52,53]
Digitalization and circular economy	Digital Twin (DT), m-IoT, hybrid AI systems—simulation and system modeling for process optimization and lifecycle management.	[43,50]
Deep learning and environmental health	CNN, deep neural networks—anomaly detection and monitoring for SDG-aligned pollution control.	[44,49]
AI and Industrial integration	CRISP-DM, clustering, predictive analytics—data-driven decision support and waste valorization in industrial contexts.	[51]

provides a concise overview of the main AI and ML techniques adopted in each research area. For each macro-theme, the table specifies the computational models, the underlying logic (e.g., regression, classification, and simulation), and the corresponding functional purposes (e.g., prediction, optimization, and monitoring). This summary helps clarify how digital methods concretely support scientific and operational advances in sustainable waste management.

This methodological breakdown reinforces the multidimensional nature of the field, showing how specific AI and ML techniques are strategically deployed to address diverse operational and environmental challenges. The integration of regression, simulation, and classification approaches across application domains highlights a growing convergence toward intelligent, adaptive, and predictive waste management systems.

4. Discussion and Conclusions

The analysis conducted in this study highlights how the application of advanced technologies such as AI and Digital Twins (DT) represents a strategic element in the transition toward more sustainable models of waste management [55,56]. The integration of digital solutions across collection, treatment, recycling, and valorization processes enables significant improvements in operational efficiency, traceability of material flows, and reductions in environmental impacts.

Beyond reiterating the key findings, it is crucial to frame them within a broader theoretical and applicative perspective. AI and DT should not be considered merely as technical tools but as enabling infrastructures for socio-technical transformation. Their integration redefines the ontology of waste management by shifting the focus from reactive to anticipatory systems, capable of adapting in real time to environmental and operational conditions.

The bibliometric analysis revealed two major research streams. The first focuses on forecasting and managing urban waste, demonstrating that predictive models based on Machine Learning (ML) enable more accurate estimations of waste generation in residential and regional contexts. This improves urban planning and reduces inefficiencies [35,45,46]. The second stream emphasizes the design of recycled and sustainable materials, showing how AI is used to optimize the mechanical and environmental properties of recycled aggregates, thus promoting more circular practices [36,40,42].

A third area concerns monitoring and environmental impacts, with a particular focus on AI and deep learning applications for real-time environmental analysis, pollutant detection, and risk assessment [41,47,51].

What emerges is not only a diversification of technological applications but a convergence of methods toward integrated, data-driven governance models. For instance, case studies such as the implementation of AI-based waste prediction models in Ho Chi Minh City (Nguyen et al., 2021 [35]) or real-time pollution detection using CNN architectures in agricultural areas of Southeast China (Chen et al., 2020 [54]) provide evidence of operational scalability and replicability.

In parallel, the digitalization of the circular economy emerges as a rapidly expanding domain. The use of DT in production and industrial environments enables continuous simulation of transformation and recovery processes, with the aim of optimizing resource flows in real time, reducing waste, and increasing systemic efficiency [42,49,56]. This approach supports the shift from a linear logic to an intelligent and adaptive management of post-consumer materials.

Another relevant area concerns the environmental management of electronic waste and heavy metals. Here, AI is used to predict contaminant accumulation in soils and water bodies, identifying environmental risk hotspots [48,49,50]. Random forest techniques and advanced spatial analysis have proven particularly effective in locating critical areas, contributing to the prevention and management of environmental damage [56,57].

From a practical standpoint, these insights point to specific policy and operational implications. Municipalities and industries should prioritize the integration of interoperable data infrastructures capable of supporting predictive maintenance, environmental surveillance, and materials tracking. For example, Reuter’s 2016 [43] framework for metallurgical process optimization via DT and m-IoT represents a prototype for sectoral transformation that can be generalized to other industrial chains.

The use of deep learning technologies for environmental monitoring also shows great potential. Fan et al., 2023 [44] and Chen et al., 2020 [54] demonstrate how the application of Convolutional Neural Networks (CNNs) and complex neural networks can improve pollutant detection, allowing for faster and more localized responses to environmental issues. These applications fit into an increasingly integrated vision linking environmental data, public health, and sustainability.

Finally, the integration of AI, industrial systems, and DT is confirmed as an emerging trend in production processes aimed at waste valorization. Fisher et al., 2020 [51] show how data-driven approaches, supported by methodological frameworks such as CRISP-DM, foster the adoption of intelligent technologies in industrial contexts, facilitating waste reduction and optimization of by-product management.

Theoretically, this convergence of digital intelligence and environmental planning supports the emergence of a new paradigm—what can be called “anticipatory circularity”—where waste is not only managed but preemptively optimized through real-time simulations and learning algorithms. This concept deserves further theoretical development and empirical validation in future research.

By integrating bibliometric mapping with thematic analysis, the study not only revealed zones of concentrated scientific activity but also uncovered emerging links between predictive methodologies, environmental monitoring, and circular economy models. In particular, the growing role of AI and DT stands out as key to the development of intelligent waste management systems capable of dynamically adapting to changes in material flows, environmental conditions, and territorial needs.

5. Study Limitations

Despite the integrated approach adopted and the richness of the sources analyzed, this study presents several methodological limitations that should be acknowledged.

First, the bibliometric analysis was conducted exclusively using the Scopus database, recognized as one of the most authoritative and comprehensive resources in the scientific landscape. However, this choice may have excluded significant contributions from other relevant databases, such as Web of Science or IEEE Xplore, especially in more interdisciplinary topics.

Second, the selection criteria primarily focused on technical-scientific domains, potentially underrepresenting contributions from social, economic, or political sciences, which could offer complementary perspectives on governance and the real-world implementation of intelligent waste management technologies.

A further limitation concerns the static nature of the analysis, which provides a snapshot of the literature within a specific timeframe but may not fully capture the rapid and continuous evolution of emerging technologies such as AI, IoT, or DT. Some recent developments may not yet be widely represented in the highly cited scientific literature.

Finally, it should be noted that this work adopts a conceptual and descriptive approach and does not include empirical validation of the technologies discussed. There is a lack of experimental data or case studies to assess the concrete impact of AI and DT solutions on actual waste management practices in urban or industrial settings.

To overcome these limitations and build upon the findings of this study, future research could follow several specific directions:

Conduct empirical case studies focused on the implementation of AI and DT systems in real waste management settings—particularly in smart cities or industrial symbiosis contexts—evaluating cost-effectiveness, scalability, and environmental performance.

Explore the integration of AI with regulatory frameworks and public policy tools to assess how predictive models and Digital Twins can support governance and decision-making in circular economy strategies.

Develop comparative analyses of algorithmic performance (e.g., ANN, CNN, RNN, and ANFIS) across different types of waste (municipal, industrial, hazardous) to identify the most effective configurations for specific applications.

Investigate the ethical, social, and organizational implications of AI and DT adoption in the waste sector, with attention to data governance, user acceptance, and workforce adaptation.

Expand bibliometric mapping to include emerging fields such as edge computing, blockchain, and generative AI in waste systems, which were outside the scope of this review but are gaining rapid momentum.

These research paths would not only address the current methodological gaps but also contribute to advancing the theoretical and operational maturity of intelligent waste management systems.