Machine Learning-Based Analysis of Community Perceptions on Coastal Forest Ecosystem Services, Restoration Willingness and Their Determinants in Bangladesh

Simple Summary

Mangrove forests in the coastal belt of Bangladesh are not only significant in protecting coastal environments from extreme events but also play a crucial role in providing various ecosystem services particularly to the communities residing nearby. Despite this, these forests areas face increasing degradation. Thus, implementing effective forest conservation strategies is essential to protect and safeguard the coastal areas and coastal communities. However, many of the conservation strategies fail to incorporate local community perspectives into the framework as community perception of mangrove ecosystem services and management remains less explored. To address this, this study utilizes machine learning models to analyze community perceptions of coastal forest ecosystem services, restoration willingness and factors influencing their perceptions. The results showed that people living close to forests supported restoration, with awareness, income, and forest dependency positively influencing their willingness. Policy recommendations include community-driven restoration approaches, enhanced environmental education, and financial incentives to encourage participation. This study offers a machine learning-informed framework for policymakers and stakeholders to design community aligned interventions. This study will pave a way to ensure the development of evidence-based strategies that ensure local participation. This will result in ensuring the long-term sustainability of restoration efforts of Bangladesh’s coastal forest.

Abstract

Coastal forests in Bangladesh provide various essential ecosystem services. However, these face severe threats from deforestation, land-use changes, and climate change. Understanding community perceptions of these services and their willingness to support restoration is crucial for effective conservation strategies. To analyze these, this study employs machine learning on survey data collected from Chakaria and Moheshkhali Upazilas of Cox’s Bazar District. Gradient Boosting Machine (GBM) model was used to identify key predictors of restoration willingness and compared with Random Forest (RF) and Generalized Linear Model (GLM). Proximity to forests shapes community priorities with stronger support for restoration among those living near the forest. Higher perception levels were linked to various ecosystem services provided by the forest, while forest dependence, age, and education also influenced perception with education and age showing strong negative correlations. The GBM model outperformed RF and the GLM in predicting restoration willingness due to its ability to capture complex relationships. The perceived importance of provisioning and supporting services and restoration awareness influenced willingness, accounting for 55.56% of the variation. Willingness was also shaped by income, forest dependency and awareness. Overall, this study leverages machine learning to align conservation efforts with socio-economic priorities, ensuring the resilience of Bangladesh’s coastal ecosystems.

1. Introduction

Coastal mangrove forests are vital ecosystems that provide numerous ecological and socio-economic benefits, particularly in tropical and subtropical regions [1]. Mangrove forests are distinguished by a blend of plant species that have adapted to the coastal environment [2]. It offers essential resources and various services that extend well beyond their immediate surroundings [3]. Globally, forests contribute to climate regulation by participating in the carbon cycle, thus helping to mitigate global warming [4,5]. They support ecological processes at the watershed level by regulating hydrological cycles, stabilizing soil, and maintaining biodiversity [6,7]. Additionally, forests are indispensable to rural livelihoods through their provisioning of food, fuel, and income-generating resources [8]. They also hold deep cultural and spiritual significance, fostering community identity and social cohesion [9]. Coastal forests, in particular, are recognized for their role in disaster risk reduction and biodiversity conservation, and for enhancing resilience to climate-induced hazards such as storm surges and sea-level rise [2,10,11]. Acting as natural buffers, mangrove forests can reduce the impact of extreme weather events like tsunamis, thereby protecting lives and property [12].

Bangladesh, with its low elevation and exposure to monsoonal weather patterns, is especially susceptible to natural disasters such as cyclones, coastal flooding, and salinization [13]. In response, the Government of Bangladesh launched a coastal afforestation program in 1966 aimed at stabilizing new alluvial lands and reducing disaster vulnerability. Over the years, mangrove plantations—featuring resilient species like Keora and Baen—have expanded significantly along the coast [14]. These plantations not only provide physical protection to coastal infrastructure but also contribute to the local economy through enhanced fishery production and employment opportunities. Despite these achievements, the country’s coastal forests are increasingly threatened by anthropogenic pressures and climate variability [15]. The sustainability of mangrove ecosystems largely depends on local community involvement in conservation and management efforts. Understanding how communities perceive ecosystem services and their willingness to participate in restoration initiatives is essential for designing effective strategies. However, these perceptions are influenced by a range of factors, including socio-demographic characteristics, economic dependency on natural resources, and environmental awareness.

Recent studies emphasize the interconnectedness of ecological and social systems within mangrove landscapes, highlighting the importance of community participation and governance structures [16]. Yet, attitudes toward mangrove ecosystem services, particularly regarding restoration and co-management, remain underexplored [17,18]. Incorporating stakeholder perspectives into mangrove governance is increasingly seen as crucial for achieving sustainable outcomes [19]. Historically viewed as contributors to environmental degradation [20], local communities are now recognized as central actors in conservation efforts [21,22]. Nonetheless, many current management frameworks fall short in integrating these diverse perspectives, often resulting in ineffective or contested interventions. Evidence from Bangladesh, Senegal, and Malaysia illustrates that acknowledging both formal and informal stakeholder interests can enhance the legitimacy and effectiveness of conservation measures [19,23,24]. Garnering broad-based stakeholder support is therefore essential to strengthening mangrove governance and restoration efforts [25,26]. Research across regions, including the South American Pacific coast and Thailand, stresses the need for tailored governance mechanisms, training, and awareness programs to ensure effective mangrove conservation [27,28]. In Thailand, community-level restoration is particularly emphasized, addressing failures linked to a lack of technical expertise [28]. Incorporating stakeholder perspectives is crucial for informed policymaking and securing necessary funding for sustainable mangrove management [29,30]. Over the years, various studies have explored the importance of ecosystem services provided by coastal forests, particularly focusing on provisioning (e.g., timber, fuelwood, and fisheries), regulating (e.g., climate regulation, flood control, and water purification), supporting (e.g., soil formation, nutrient cycling, and habitat provision), and cultural services (e.g., recreation, spiritual value, and traditional knowledge) [27,31,32,33,34,35]. Many studies have emphasized the role of socio-demographic factors such as education level, income, age, and occupational dependence on forest resources in shaping public attitudes toward conservation and restoration [28,36,37,38,39]. Traditional statistical methods like regression analysis and factor analysis have been widely used to assess these relationships [40,41,42,43]. However, these approaches often assume linearity and may not adequately capture the complex, non-linear, and interactive effects among multiple influencing factors [44]. Recent advances in machine learning (ML) have provided new opportunities to overcome these limitations by allowing more flexible, data-driven analysis of large and complex datasets. Machine learning techniques such as Gradient Boosting Machine (GBM), Random Forest (RF), and Generalized Linear Model (GLM) have demonstrated higher predictive accuracy and better feature selection capabilities in socio-ecological studies [45,46]. While machine learning (ML) has been increasingly adopted in environmental research for tasks such as remote sensing analysis, species habitat modeling, and climate prediction, its application in the social dimensions of environmental governance particularly in community perception and behavior studies remains relatively limited. Most existing socio-environmental studies still rely on linear regression-based models, which may oversimplify complex human–nature interactions by failing to account for non-linear and interacting effects among multiple variables. The present study contributes to this growing but still nascent field by applying ML models not only to classify community willingness to support restoration but also to interpret the relative importance of socio-demographic and perceptual drivers in the context of Bangladesh.

Despite valuable contributions from previous research, significant gaps remain in the integration of advanced analytical tools to assess community engagement in coastal forest restoration. Most existing studies in tropical or mangrove contexts have relied on conventional statistical approaches such as correlation or logistic regression, which typically assume linear relationships and may not adequately capture the complex, non-linear interactions between socio-demographic factors, ecosystem service perceptions, and restoration behavior. For instance, Amone-Mabuto et al. [47] assessed seagrass ecosystem perceptions in Mozambique through interviews and surveys, emphasizing fisheries as key services, but did not examine behavioral responses such as restoration support. Yetein et al. [48] used indices and logistic regression to relate land-use types to perceived ecosystem services in Benin but focused solely on descriptive associations. Zeratsion et al. [49] analyzed urban forest perceptions in Ethiopia using regression models but did not address willingness to engage in restoration. Studies by Dat et al. [50] and Pham et al. [51] examined willingness to pay for mangrove services in Vietnam using PLS-SEM and contingent valuation, respectively, identifying socio-economic predictors but without integrating broader ecosystem service perceptions. These limitations hinder a comprehensive understanding of how multiple variables jointly influence community support for ecological restoration. Our study offers a novel contribution by integrating ecosystem service perceptions with restoration willingness using machine learning (ML) models such as Gradient Boosting Machine (GBM), Random Forest (RF), and Generalized Linear Model (GLM), an approach not previously applied in this context. Unlike traditional statistical methods, ML techniques can capture complex, non-linear interactions within high-dimensional data. This methodological advancement enables more nuanced and accurate classification of behavioral responses and identification of key predictors. As such, this study provides a data-driven foundation for participatory planning and informed policy design in coastal forest restoration efforts in Bangladesh.

Moreover, research on coastal forest restoration in Bangladesh has predominantly focused on ecological and biophysical aspects, often overlooking the critical role of community perceptions in shaping conservation success [52,53,54]. In addition to that, mangrove forests in Bangladesh have significantly declined over the years due to population growth and livelihood dependence, with the Chakaria Sundarbans losing 87.5% of its coverage between 1972 and 2017 due to human activities like shrimp farming and salt mining [55]. As one of the most climate-vulnerable countries, Bangladesh hosts the largest contiguous mangrove forest, yet research on the societal and environmental benefits of restored mangroves remains scarce. Understanding community perceptions of mangrove restoration is crucial for effective conservation efforts. Therefore, this study aims to bridge these gaps by employing machine learning models to analyze community perceptions of coastal forest ecosystem services, restoration willingness, and their determinants, offering a data-driven perspective on coastal forest conservation. Specifically, this study seeks to answer the following research questions:

• How do coastal communities perceive different ecosystem services (provisioning, regulating, cultural, and supporting) provided by coastal forests?
• What socio-demographic and perception-related factors influence the community’s willingness to support coastal forest restoration activities?
• Can machine learning models effectively classify and predict community willingness to support restoration based on socio-economic and perception variables?

The findings of this study provide actionable insights for policymakers and forest managers. By identifying the specific ecosystem services most valued by local communities, and understanding the factors driving restoration support, the research supports the development of targeted, community-based conservation strategies. Moreover, the use of machine learning enhances the capacity for data-driven environmental planning, offering scalable methods for prioritizing areas and groups for engagement in restoration efforts across similarly vulnerable coastal zones.

2. Materials and Methods

2.1. Study Area

The study area comprises Chakaria and Moheshkhali Upazila on the southeastern coast of Bangladesh (Figure 1). Geographically, Chakaria Upazila stretches between approximately 21°15′ N to 22° N latitude and 91°45′ E to 92°15′ E longitude, with the Matamuhuri River as the main waterway, intertwined with numerous canals, tributaries, and micro-channels [56]. Moheshkhali Upazila which is an Island, positioned within 21°20′ N to 21°50′ N latitude and 91°45′ E to 92°00′ E longitude, is bordered by Chakaria and Cox’s Bazar to the north, northeast, east, and southeast, with the Moheshkhali channel running across it [57]. Its geological complexity is marked by hilly topography surrounded by coastal plains, exhibiting distinct geological, tectonic, and geomorphological features. Moheshkhali Island’s landforms are intricately linked to its geological deposition processes, divided into subdivisions including active, young, and old coastal plains, as well as hilly areas with piedmont plains. Notably, it experiences two distinct depositional systems: closed to semi-closed in the south and open to semi-closed in the north. The island is undergoing an accretionary phase, expanding at a rate of approximately 1.2 square kilometers per year since 1972 [57]. Recently, the study areas have faced climate change threats, habitat degradation, and coastal erosion, making conservation efforts essential. Therefore, evaluating community perceptions of ecosystem services and willingness to participate in restoration in Chakaria and Moheshkhali Upazila is crucial due to their ecological sensitivity and socio-economic dependence on coastal resources. Moreover, understanding how communities value provisioning, regulating, supporting, and cultural services helps design effective, community-driven restoration programs. Ultimately, this research supports coastal ecosystem resilience, sustainable resource management, and climate adaptation, ensuring conservation aligns with both ecological priorities and local needs.

2.2. Research Design, Questionnaire Development, and Data Collection

A mixed-methods approach was employed, combining both qualitative and quantitative techniques to ensure comprehensive understanding of community perceptions and restoration willingness [54]. This research was conducted by a multidisciplinary team comprising environmental scientists and trained field enumerators. First, the survey instrument was designed based on an extensive review of relevant literature focusing on ecosystem services, community perception, and restoration willingness in coastal and forested areas. The draft questionnaire was then refined through consultations with academic experts, local forestry officials, and development practitioners to ensure contextual appropriateness and technical accuracy. To validate the relevance and clarity of the questionnaire items, two focus group discussions (FGDs) were conducted. Participants were purposefully selected based on their interactions with the coastal forest, such as collecting forest products, fishing, or participating in conservation activities. These FGDs helped identify 17 ecosystem services relevant to local communities and informed the construction of perception and willingness indicators for the questionnaire (

Table 1. Ecosystem services identified through FGDs and perceived importance by coastal communities.

Ecosystem Service Category	Ecosystem Service Identified	Perceived Importance	Key Discussion Highlights from FGDs
Provisioning Services	Timber	Moderate	Valued more by distant communities for construction; lesser importance to forest-dependent households
	Food (Fisheries)	High	Frequently cited by fishing households as critical for sustenance and income
	Water	Moderate	Recognized as vital for agriculture and domestic use, especially during dry seasons
	Fuelwood	High	Most cited provisioning benefit by both men and women, essential for cooking and heating
Regulating Services	Climate Regulation	High	Linked to temperature stability and seasonal rainfall regulation
	Flood Control	High	Mentioned in both FGDs as critical, especially during monsoon season
	Erosion Protection	High	Strong concern over riverbank and shoreline erosion; forests seen as natural barriers
	Protection from Storms	Very High	Cited as most crucial regulating service, especially post-cyclone experiences
	Protection from Tidal Surges	Very High	Participants emphasized that coastal forests act as natural barriers, reducing the impact of tidal surges during cyclones and high tides
Cultural Services	Tourism Opportunities	Moderate to High	Identified as emerging economic opportunity, especially by younger participants
	Recreation Opportunities	Moderate	Valued by youth; adults viewed recreation as less relevant to daily life
	Esthetic Value	High	Expressed appreciation for scenic beauty, peace, and well-being from forest environment
	Education Opportunities	Moderate	Seen as a source of school excursions and environmental learning
	Research Opportunities	Low to Moderate	Recognized by few participants; some mentioned visits by researchers but did not directly benefit
Supporting Services	Habitat for Animals	High	Linked to forest biodiversity and hunting (by some); appreciated for ecological value
	Habitat for Birds	Very High	Widely valued for bird watching and cultural symbolism
	Habitat for Fisheries	High	Connected to nearshore fishing activities and spawning grounds

).

Prior to the full-scale survey, a pilot test was conducted with 15 households drawn from similar socio-ecological contexts but not included in the final sample. The goal was to assess the clarity, relevance, and internal consistency of the questionnaire items derived from the 17 ecosystem services identified through focus group discussions. Based on the pre-test, minor wording revisions were made to improve question comprehension and reduce ambiguity, particularly in items related to cultural and supporting services. The Likert-scale structure (1 = Not Important to 5 = Very Important), used to measure perceived ecosystem service importance and restoration willingness, was confirmed as understandable and appropriate for the target population. To assess internal consistency, Cronbach’s alpha was calculated for grouped perception items within each ecosystem service category (provisioning, regulating, cultural, and supporting). The reliability scores ranged from 0.76 to 0.85, indicating acceptable to good internal consistency [58]. These results confirmed that the questionnaire items were coherent and capable of reliably measuring the intended constructs across community groups.

Fieldwork was conducted from 5th to 25th March 2024, in Chakaria and Moheshkhali Upazilas. Key informant interviews (KIIs) were conducted with local forest officers, Union Parishad members, and NGO representatives working in forest conservation. These interviews were used to validate community-reported information and gather expert opinions on forest degradation and restoration interventions. Field observations were carried out to assess environmental features (e.g., visible erosion, afforestation signs, and forest use), which complemented community narratives. Observations were particularly useful in verifying ecosystem service use and understanding contextual differences across sites. For the quantitative survey, household heads or adult decision-makers were targeted to ensure informed responses regarding socio-economic status, forest use, and attitudes toward restoration. Prior to participation, respondents were informed about the study’s objectives, and verbal consent was obtained. Data collection was carried out by trained field enumerators who received instruction on the research objectives, ethical guidelines, and standardized protocols for survey administration. The structured interviews were conducted in Bengali to ensure clear communication with respondents.

2.3. Sample Size and Sampling Procedure

While the general population of Chakaria and Moheshkhali exceeds several hundred thousand, this study specifically targeted a subpopulation of forest-dependent households—those directly or indirectly relying on coastal forest resources for livelihoods, safety, or cultural value. As such, the sampling frame was constrained to forest-adjacent zones rather than the broader population.

In this study, the sample consisted of community members with varying levels of interaction with coastal forests in Chakaria and Moheshkhali Upazilas of Cox’s Bazar District. The general population across both Upazilas exceeds several hundred thousand; however, the target population for this study was more narrowly defined to include only those households with direct or indirect dependence on coastal forest resources. The sampling approach was thus focused, not random, across the general population, but purposive within forest-adjacent zones. The sample size was determined using a widely accepted formula for estimating proportions in population studies [59,60,61] using the following equation:

$$Sample Size={\displaystyle \frac{z^2\times Þ(1-Þ)}{e^2}}$$

where z is the z-score (1.96 for 95% confidence), p is the estimated population proportion (0.5 for maximum variability), and e is the margin of error (0.08). Applying this yielded a sample size of 151, which was operationally rounded to 150 for field implementation.

Due to logistical challenges, including the dispersed settlement patterns and transportation constraints in coastal villages, a total of 150 respondents (75 from each Upazila) were selected using simple random sampling within the identified forest-interacting zones. The emphasis was on representativeness of forest users rather than statistical generalization to the full district population. This allowed us to gather in-depth insights aligned with the study’s objectives. Furthermore, household heads or adult decision-makers were prioritized as respondents to ensure informed responses regarding ecosystem service use and restoration support. Our focus was on households that interacted to some extent with coastal forest, rather than the entire village population. The sample included a higher proportion of male-headed households, reflecting traditional gender roles in the study area. Women’s participation in public discussions and decision-making processes was limited, and their reluctance to venture deep into coastal forests also influenced the sample composition.

2.4. Variables of the Study and Their Measurement

The variables for the final survey were divided into three sections in this study. The first section included in the questionnaire was about socio-demographic characteristics of the respondents. These variables include age, level of education, duration of the residency period, distance from coastal forest, annual family income, and households’ dependency on coastal forest resources. Among these variables, age, duration of the residency period, and education were measured in years while distance from coastal forest in kilometers, monthly income in BDT (Bangladeshi currency, i.e., Taka), and households’ dependency on coastal forest resources in a 6-point rating scale. The second section encompassed the variables related to the respondents’ perceived importance of coastal forest ecosystem services. To determine the perception on ecosystem services the respondents were left free to perceive the importance ecosystem services based on their own understanding and values as similar to the other studies conducted by Wardropper et al. [62] and Su and Gasparatos [26]. In this study 12 ecosystem services were selected under four categories, i.e., provisional, regulating, cultural, and supporting ecosystem services. Following an extensive literature review the provisional ecosystem services included food production, fuelwood, water, and timber [26,63,64]. Regulating services comprised climate regulation, flood control, erosion protection, protection from tidal surge, and protection from storm [34]. Cultural ecosystem encompassed tourism, recreational, esthetics, education, and research [45]. Supporting services included perceived importance of habitat for animals, habitat for birds, and habitat for fisheries [26,65]. All the variables related to ecosystem services were measured in a 5-pont Likert scale. In the third section, variables related to perception of coastal forest restoration were explored, namely ongoing awareness of restoration initiatives and willingness to participate in future restoration initiatives. Awareness of restoration was measured as a binary response of the respondents, i.e., ‘yes’ or ‘no’, while willingness to participate in the restoration activities was measured in a 5-pont Likert scale.

2.5. Statistical Analysis and Analytical Framework

Data collected using the structured questionnaire first stored in the Microsoft Excel spreadsheet for data cleaning, and standardization to ensure comparability. Continuous variables such as age, income, education level, residency period, income, and distance to coastal forest were normalized, while categorical variables were transformed using dummy encoding where necessary. After that the data were imported to the SPSS (V30.0.0) and R-software (version 4.2.2) for further analysis. To characterize the socio-demographic characteristics of the respondents, simple statistical measures such as mean, standard deviation, and frequency distribution were used. This study was designed to test the following hypotheses:

Hypothesis 1:

Socio-demographic variables and the perceived importance of ecosystem services are positively associated with the willingness to support restoration activities.

Hypothesis 2:

Machine learning models can accurately classify and predict willingness to support restoration based on combined socio-demographic and perception data.

To test Hypothesis 1, we used Principal Component Analysis (PCA) to identify the significant variables affecting respondents’ perceptions of ecosystem services. To test Hypothesis 2, we employed three supervised classification algorithms, Gradient Boosting Machine (GBM), Random Forest (RF), and Generalized Linear Model (GLM), to predict community willingness to support restoration activities. These machine learning models were applied to explore the relationship between socio-demographic factors, the perceived importance of ecosystem services, and respondents’ willingness to support restoration efforts.

2.5.1. Principal Component Analysis (PCA)

PCA, a dimensionality reduction statistical technique, captures the maximum variance in the data while preserving the most significant patterns in a dataset. It transforms a set of correlated variables into a smaller number of uncorrelated variables used in research and data analysis to simplify complex datasets reduce redundancy, and enhance interpretability [66]. We used PCA to identify the most important factors affecting the community’s perception of ecosystem services because PCA identifies principal components (PCs), linear combinations of the original variables, that capture the maximum variance in the dataset. These components reveal which factors contribute most to shaping perception [67,68]. Moreover, PCA assigns factor loadings to each variable, indicating how strongly each variable contributes to a principal component. Furthermore, variables with high absolute factor loadings in the first few components are considered most influential in explaining variations in community perceptions. In this study, the relationship between the relative importance of ecosystem services and community groups based on their distance to the coastal forest was first explored. Then, the relationship between socio-economic characteristics, the relative importance of ecosystem services, and the overall perception level of community groups was investigated using PCA. The PCA was carried out in R software (version 4.2) using the FactoMineR and factoextra packages [69].

2.5.2. Machine Learning Model

Machine learning models were used to explore the relationship between socio-demographic factors, the perceived importance of ecosystem services, and willingness to support restoration activities. Three models were employed: Gradient Boosting Machine (GBM), an ensemble learning technique that builds models sequentially, correcting previous errors to improve accuracy and handle complex relationships [70]; Random Forest (RF), a tree-based ensemble method that constructs multiple decision trees and aggregates their outputs to enhance prediction accuracy and reduce overfitting; and Generalized Linear Model (GLM), an extension of linear regression that accommodates different response distributions and uses a link function to model relationships between variables, making it useful for interpreting predictor impacts [46,70,71,72].

Data was collected through structured questionnaires targeting local communities near coastal forests, covering socio-demographic factors (age, gender, education, income, and dependency on forest resources), the perceived importance of ecosystem services (provisioning, regulating, supporting, and cultural), and willingness to support restoration activities. Before applying machine learning models, the data was preprocessed by handling missing values, encoding categorical variables, and standardizing continuous variables.

In this study, willingness to participate in restoration activities, a dichotomous variable, was used as the response variable, while socio-demographic factors and the perceived importance of ecosystem services were used as predictor variables to classify respondents based on their perceptions. The dataset was split into training (80%) and testing (20%) sets, with hyperparameter tuning performed using cross-validation. Model performance was evaluated using accuracy, precision, RMSE (Root Mean Square Error), log loss, and AUC-ROC (Area Under the Receiver Operating Characteristic Curve). After evaluating the machine learning models, one model was selected for this study. Based on the selected model, feature importance was assessed, and a partial dependency plot was created to identify the factors influencing respondents’ perceptions and how a unit change in predictor variables affects perception levels. In this study, R software was used for implementing machine learning models, and all analyses were performed using the H2O package [73].

3. Results

3.1. Relative Importance of Ecosystem Services Provided by Coastal Forest According to Respondent’s Perception

In this study, 17 ecosystem services were identified for the coastal forest, and these were categorized into four major ecosystem services, i.e., provisioning (4), regulating (5), cultural (5), and supporting (3) services. These ecosystem services were identified through a focus group discussion (FGD) for both the communities (n = 20, for each community) who are close to the forest and far from the forest. The community’s perception was compared across four major ecosystem services using spider diagram analysis presented in Figure 2. Among the provision services, communities who are close to the coastal forest showed high perceived importance in terms of food (PIF), water (PIW), and fuelwood (PIFW) compared to the communities who are living far from the coastal forest. On the other hand, communities that are living far from the coastal forest considered timber (PIT) as the most important provisional service (Figure 2a). Though 5-point Likert scale distribution of provisional services varies among the two community groups (Figure 3a), overall, 58.1% of residents perceived fuelwood as very high important (Figure S1). In addition, provisional services, such as food and fuelwood, were considered high to very high in importance by 80% of the respondents who are close to the forest compared to the other services (Figure 3a). Survey data revealed that both communities of respondents placed high importance on various regulating benefits such as climate, flood control, erosion protection, protection from storms, and protection from tidal surges (Figure 2b). Among the regulating services, protection from tidal surge (PITS) was perceived as the most significant service by both groups compared to the other regulating services (Figure 2b). Communities living close to the forest ranked climate regulation, protection from floods, and erosion control as noteworthy regulating services. Overall, the protection from tidal surge and protection from erosion ranked very high in importance by the respondents, 89.42% and 83.65%, respectively (Figure S2), and it was almost 100% for both groups (Figure 3b). In terms of cultural service, respondents close to the coastal forest had a more positive perception compared to the other groups. Esthetic value of the coastal area was identified as the most vital cultural service followed by tourism by the groups living close to the forest (Figure 2c). In summary, esthetic services were identified as high to very high in importance for cultural services by 62.8% of the respondents, whereas it was 68% for the groups living close to the coastal forest (Figure 3c). Furthermore, recreation and tourism cultural services are perceived as high to very high by 59% of respondents who are residing near a coastal forest (Figure S3). In the case of supporting services, more than 80% of the respondents of the both groups showed positive perception (Figure 2d) and assigned high to very high importance to three supporting services (Figure 3d); however, overall, 96.19% of respondents ranked habitats for birds as high to very high in importance as supporting services by the communities (Figure S4).

3.2. Factors Affecting Communities’ Perception of Ecosystem Services

In this study, to evaluate the relationship between socio-economic factors, the relative importance of ecosystem services, and the perception level of community groups, principal component analysis was performed. Figure 4 illustrates the relationship of relative importance of ecosystem services and groups of communities who are living close to and far from coastal forests. The results presented in Figure 4 indicate that, collectively, 40.7% of the variance explained by the first two dimensions showed that perceived ecosystem importance was directly associated with two groups of the respondents. Among the two dimensions, dimension-1 (Dim1) contributed most, which accounts for 27.2% of the variance compared to dimension-2 (Dim2). Dim1 represents most of the community members who are living close to the coastal forest, which is positively correlated with the perceived importance of all provisional, regulating, cultural, and supporting ecosystem services. While all of these ecosystem services are important, cultural (including recreation, tourism, and esthetics; 12.26%, 9.23%, and 9.22%, respectively), supporting (including bird habitats; 9.08%), and regulating (including flood control and erosion protection; 8.87% and 8.02%, respectively) ecosystems are the most important. These ecosystems also have a strong positive relationship with community groups. However, Dim2 represents most of the community that is living far from the coastal forest, which is positively correlated with the perceived importance of regulating, supporting, and provisional services, while only cultural services are negatively correlated. Among the positive associations, the protection of ecosystem services from tidal surges (8.27%) and protection from storms (7.67%) contributed significantly. In the biplot analysis, the highest loading factors were found to be the importance of recreation (0.753) for Dim1 and education (−0.688) for Dim2.

Figure 5 represents the association of socio-economic and relative importance of ecosystem services with the perception level of the community. Findings contained in Figure 5 reveals that, jointly, 37.1% of the variance explained by the first two dimension showed that socio-economic variables and perceived ecosystem importance directly corelated with the perception level of the respondents. Dimension-1 (Dim1) represents mostly high and medium perception level respondents, explaining 21.1% of the variance. All the variables showed a positive relationship, with the exception of the respondents’ age and income, as shown in Figure 1 for Dim1. The factors that affected the respondents’ perception level were perceived regulating, provisional, supporting, cultural ecosystem services, and dependence on forest resources. These factors contributed significantly, accounting for 28.53%, 23.99%, 19.68%, 18.88%, and 5.15% of the contribution to the perception level. Dimension-2 (Dim2) contains the respondents with low and medium perception level, explaining 16% of the variance. Socio-economic factors like education, age, dependence on forest resources, and how important coastal forests are all have a big impact on the respondents’ perception level. Education (41.1%) and age (21.13%) were the main factors, with education having a negative relationship with the perception level.

3.3. Willingness of the Community to Support Restoration Activity

This research makes an effort to investigate how likely respondents are to be willing to participate in restoration activities for improved ecosystem services. We first asked the respondents if they were aware of any ongoing afforestation initiatives in their community, and Figure 6a presents this information. More than three-fifths (76%) of respondents who are close to the coastal forest were aware of forest restoration in their community (Figure 6a), and 85% of this community are very willing to extremely willing to support restoration activities in the coastal area (Figure 6b). Meanwhile, 62% of the respondents who were aware of forest restoration living far from coastal forests (Figure 6b) responded that 83% of them were very willing to extremely willing to participate in restoration activities. From the results, it is evident that over 80% of the respondents in both groups are very willing to extremely willing to support restoration activities, as shown in Figure 6b. Nevertheless, overall, a total of 84% of respondents classified themselves as very willing to extremely willing to support restoration activities (Figure 6c).

3.4. Factors Influencing Communities’ Willingness to Support Restoration Activities in the Coastal Areas

Nine socio-economic characteristics were selected as response variables to classify the community as “yes” on whether they perceived that they were willing to support restoration activities (here we considered moderate to extremely willing categories for “yes”) and “no” on whether they were not willing to support restoration activities in the future in the coastal area (here we considered slightly to not willing categories for “no”). The descriptive statistics of the predictor’s variables are presented in

Table 2. Socio-economic characteristics of the respondents.

Characteristics	Range or Categories	$\mathbf{Mean} \pm$ SD for Continuous or Percentage of Farmers for Categorical Variables
		“Yes” Perception *	“No” Perception *
Age (years)	18–82	$39.59 \pm$ 14.83	$41.38 \pm$ 11.26
Residency period (years)	2–70	$33.50 \pm$ 14.40	$38.02 \pm$$14.16$
Monthly income (BDT)	2000–50,000	$17,424.52 \pm$ 9697.27	$17,818.18 \pm$ 8957.97
Perceived importance of provisional services (Score)	6–20	$14.19 \pm$ 2.98	$14.68 \pm$ 2.93
Perceived importance of regulating services (Score)	13–25	$21.83 \pm$ 3.57	$20.95 \pm$ 3.85
Perceived importance of cultural services (Score)	9–15	$13.83 \pm$ 1.69	$12.31 \pm$ 5.15
Perceived importance of supporting services (Score)	5–25	$14.42 \pm$ 5.57	$12.88 \pm$ 1.81
Distance from coastal forest (Km)	1–10	$4.85 \pm$ 3.35	$4.47 \pm$ 2.61
Education (years)	No education	38.7	52.2
	Primary education	35.8	22.7
	Secondary education	19.8	20.5
	Higher secondary	3.8	2.3
	Graduate	1.9	2.3
Dependency on coastal forest	Not dependent	14.2	25.0
	Somewhat dependent	11.3	9.1
	Moderately dependent	30.2	29.5
	Highly dependent	24.5	20.5
	Very highly dependent	19.8	15.9
Importance of coastal forest	Not important	-	2.3
	Moderate importance	9.4	2.3
	High importance	32.1	38.6
	Very high importance	58.5	56.8

* “Yes” indicates moderate to extremely willing categories and “No” indicates slightly to not willing categories.

. The results presented in Table 1 indicate that communities that were willing to support restoration activities had more dependency (which was 19.8% of the respondents for very high dependency category) on forest resources than those that were not willing to support restoration activities. Those who depended on forest resources more and were willing to support restoration activities also had a more positive view of the importance of different ecosystem services, such as supporting services (mean score: 14.42), cultural services (mean score: 13.83), and regulating services (mean score: 21.83). This was in contrast to the groups that were not willing to support restoration activities. To understand the relative influence of these factors on willingness to support the restoration activities, we used three machine learning approaches, such as Random Forest (RF), Gradient Boosting Machine (GBM), and General Linear Modeling (GLB). First, the performance of these three models was checked and evaluated in terms of RMSE, accuracy, log loss, precision, and AUC to identify the predicting ability of the machine learning model.

The results presented in

Table 3. Performance of the models in the training dataset.

Models	RMSE	Accuracy	Logloss	Precision	AUC
GLM	0.39	0.64	0.40	0.66	0.61
RF	0.35	0.670	0.37	0.75	0.72
GBM	0.11	0.77	0.12	0.77	0.70

reveal that GBM outperforms the other two models in terms of low RMSE and log loss compared to the other two models, such as RF and GLM, which have high RMSE and log loss. Moreover, GMB models produced a good precision rate (0.77) and AUC (0.72) value (Figure S5); in turn, this model demonstrated high accuracy (77), indicating a good prediction ability to classify whether the respondents would support restoration activities or not. The residual distribution of the three models implies that the GBM model appears to perform better than the other two models, given the median residuals, which are lowest for the GBM model with a lower number of tail residuals (Figure S6). We further validated the GMB model using 10-fold cross-validation metrics. The results of the cross-validation matrices suggested a good performance of the GBM model with high accuracy (ranges from 74 to 83%) and AUC values (ranges from 72 to 83%) in the cross-validation dataset (Figure S7). Finally, we selected the GBM learning models to classify the communities into two groups: those who were willing to support the restoration activities and those who were not willing to support the restoration activities. We ran a simulation of variable importance using 8000 samples to find out how each predictor variable affected how willing people in the community were to support restoration activities on average. Among the 12 features, all are common to both the RF and GBM models, with distance from coastal forest being the single most important predictor in all of them (Figure S8). The variable importance plot (Figure 7a) from the GBM model shows that the first three factors—distance from coastal forest, perceived importance of supporting services, awareness on forest restoration made up 55.56% (Table S1) of the overall perceived willingness to support restoration activities near the coast or not. The remaining nine variables influenced the overall perception by 44.44%. We also investigated the relationship of the predictor’s variables with the predicted probability of willingness to support restoration activities using partial plot (Figure 7b). The partial plot demonstrated the non-linear relationship between the predictor variables and the respondents’ perceptions of their willingness to support restoration activities. The respondents’ perception of the importance of supporting services, their awareness of forest restoration, and their income all showed a positive correlation with their perceived willingness to support restoration activities. The education level of the respondents negatively affected the respondent’s perception of forest restoration almost in a linear fashion. Among the predictors, the respondent’s age exhibited significant variability in predicting the likelihood of supporting restoration activities. The respondents with younger ages had a more positive perception of supporting restoration activities, while willingness sharply declined with the increase in age and reached a lowest peak at 45 years; after that, it increased for the communities over 50 years old. The respondents with the increases in dependency on forest resources, the probability of willingness to support forest resources also increases, but after that, it slightly declines and remains steady though the respondent dependency increases. The probability of willingness to support restoration activities initially decreases as the distance from the coastal forest increases. However, beyond a certain point, it shows a positive relationship as communities farther from the forest exhibit greater willingness to support restoration efforts.

4. Discussion

Coastal forest restoration is critical for sustaining ecosystem services and enhancing the resilience of coastal communities. As these services underpin ecological stability and human well-being, understanding how they are perceived by local populations is essential for designing socially acceptable and ecologically effective restoration strategies [74,75,76,77]. This discussion integrates findings on community perceptions of the importance of ecosystem services and restoration activities, examining how these perceptions are influenced by socio-economic factors, as presented in the following subsections. Furthermore, it outlines key policy implications, acknowledges methodological limitations, and provides directions for future research.

4.1. Perceived Relative Importance of Ecosystem Services and Willingness to Support Restoration Activities

The coastal communities in the Cox’s Bazar district of Bangladesh prioritize a diverse array of forest ecosystem services, as demonstrated by our investigation. Forest ecosystem services are essential for the cultural and economic well-being of those communities that are more or less reliant on forest resources, as well as for their livelihood in Bangladesh [78,79]. Our findings reveal that coastal communities prioritize ecosystem services differently based on their proximity to forested areas. Those living near coastal forests place greater importance on provisioning services such as food, water, and fuelwood reflecting their direct dependence on these resources. Fuelwood, in particular, was consistently rated highly due to its dual role as a primary energy source and income generator, aligning with previous studies [34,78,80]. Regulating services like erosion control, flood protection, and climate regulation were widely recognized, especially among communities most exposed to natural disasters. Interestingly, storm protection was rated more highly by distant communities, possibly due to their indirect experience of forests as barriers, while nearby residents, often relocated during storms, may not associate forests with immediate protection. This finding has already been confirmed by Hossain et al. [81] and Bhowmik et al. [82] in their study. Cultural services were appreciated most by those in close proximity to the forest, particularly for their esthetic and recreational value. Esthetic appreciation, as supported by prior research [83,84] contributed significantly to emotional and psychological well-being. In contrast, cultural services were less valued by distant communities, likely due to limited access and reduced experiential interaction. This is consistent with human inclinations for visually pleasing and serene situations that promote well-being and foster a deep emotional connection with nature [84,85]. Tourism was the second most significant service, most likely because the visual features of coastal forest entice tourists and give chances for leisure, relaxation, and economic advantages for local communities via tourism activities [86,87]. Supporting services, including habitat provision for birds, fish, and animals, were highly valued across both community groups, reflecting strong awareness of the forests’ biodiversity and ecological importance. Communities perceived coastal forests as vital habitats due to their rich biodiversity, abundant food sources, and safe breeding grounds. The forests’ dense vegetation and submerged roots support wildlife, regulate water quality, buffer extreme weather, and connect terrestrial and aquatic ecosystems, contributing to ecological stability and favorable microclimates [88,89,90].

Regarding the communities’ perception on willingness to support restoration activities, both groups showed positive perception, though those living close to the coastal forest had more perception on willingness to support restoration activities. The ecosystem services that coastal forests provide directly benefit the communities in close proximity, which is why they are more inclined to support restoration activities. These consist of enhanced water quality, erosion control, protection against cyclones, and access to resources such as timber and fish. The emotive and cultural connection to the region is further bolstered by the esthetic and recreational value of the forests [91]. Furthermore, these communities are more susceptible to the adverse consequences of forest degradation, including the extinction of biodiversity and the destruction of livelihoods. Consequently, they are more inclined to take action in order to safeguard and restore these essential ecosystems.

4.2. Factors Affecting Perception of Ecosystem Services and Willingness to Support Restoration Activities

Principal Component Analysis (PCA) and machine learning models revealed that provisioning, supporting, and cultural services were key predictors of ecosystem service perception. Communities in close proximity to the coastal forest, and dependent on its resources, exhibited higher perception scores likely due to the direct, tangible benefits they receive. Their livelihood is sustained by provisioning services, including fish, timber, and medicinal plants. These findings are consistent with findings of Barua et al. [34] in their perspective studies. Their economic opportunities and well-being are improved by cultural services, such as esthetic value, tourism, and recreation [92]. In contrast, cultural services showed a negative correlation with distance, emphasizing the role of physical proximity in appreciating experiential benefits. Communities living farther from the coastal forest have fewer opportunities to access or engage with these services, resulting in lower perceived importance. Since cultural services do not offer immediate survival benefits, they are often undervalued compared to essential resources [93,94], which explains why their perceived importance decreases with increasing distance. All variables, including regulating, provisioning, supporting, cultural ecosystem services, and dependence on forest resources, positively influence respondents’ perception of ecosystem services because they directly benefit from these functions. However, age and education are exceptions to this positive relationship. Older individuals may have different experiences or reduced direct interaction with forests, leading to varying perceptions [95]. Similarly, higher education levels may provide alternative livelihoods, making people less dependent on forest resources and prioritizing development over direct ecosystem benefits. As a result, while most factors enhance positive perceptions, age, and education show weaker or different associations.

Among the machine learning models tested, the Gradient Boosting Machine (GBM) model outperformed Random Forest (RF) and Generalized Linear Model (GLM) in predicting restoration willingness, offering robust accuracy and sensitivity to non-linear relationships. The GBM model outperformed the other models due to its ability to capture complex, non-linear relationships and assign higher importance to influential features by adjusting weights, reducing bias towards dominant classes [96,97,98]. These advantages make GBM a more precise and robust model for predicting community perception toward restoration support. The top three factors, perceived importance of provisioning services, awareness of forest restoration, and perceived importance of supporting services, accounted for 55.56% of the overall willingness to support restoration because they directly shape how communities value coastal forests and their motivation to engage in conservation. Moreover, these variables reflect both material dependence and ecological awareness as drivers of restoration support. Together, these three factors strongly influence perceptions, emphasizing the need for awareness programs and policies that highlight both direct and indirect benefits of coastal forest restoration.

Partial dependency plots revealed that support services, forest restoration awareness, and income positively influenced willingness to support restoration, as individuals with ecological knowledge or stable incomes are more likely to engage in conservation. In contrast, education showed a negative linear relationship, possibly reflecting a shift in priorities toward economic growth. Age had a non-linear effect, with greater willingness among younger and older individuals and a dip in midlife, likely due to shifting personal and economic priorities [99]. Forest dependency also showed a positive association, though willingness plateaued at higher dependency levels, possibly due to concerns over future access restrictions [26].

4.3. Implications of the Research for Policy Formulation

This study provides actionable insights for policymakers aiming to design inclusive and sustainable restoration strategies. First, interventions should be spatially targeted: communities near forests should be engaged through participatory restoration programs, while those farther away may benefit more from awareness campaigns emphasizing protective and ecological benefits. Second, restoration programs should align with livelihood incentives. Provisioning and supporting services are strongly linked to restoration willingness; thus, integrating agroforestry, ecotourism, and sustainable harvesting into restoration schemes can reinforce community buy-in. Third, demographic targeting can enhance engagement. Programs should focus on youth and older adults who demonstrated higher restoration support. Education initiatives should also address the knowledge–action gap among educated individuals by highlighting the socio-ecological value of forests.

On a broader scale, the findings of this study underscore the value of machine learning as a tool for advancing socio-environmental research. The ability to classify and predict community restoration willingness enhances the efficiency and responsiveness of conservation planning. Similar approaches can be replicated in other coastal or mangrove-dependent regions facing deforestation and climate threats.

4.4. Limitations of This Research and Future Research Directions

While this study offers valuable insights into community perceptions of ecosystem services and willingness to support restoration using machine learning models, several methodological and analytical limitations should be acknowledged. The sample size of 150 households, although statistically justified for this study’s objectives, may not capture the full diversity of perceptions across all coastal regions in Bangladesh. The focus on two Upazilas could introduce geographic or contextual biases, and findings should thus be interpreted as context-specific rather than broadly generalizable. In addition, potential response biases—such as social desirability or misinterpretation of questions due to variations in literacy or familiarity with environmental concepts—could affect the accuracy of self-reported data. Although the questionnaire was pre-tested for clarity, relying primarily on household heads may have limited the diversity of viewpoints, particularly those of women or younger family members.

The machine learning models Gradient Boosting Machine (GBM), Random Forest (RF), and Generalized Linear Model (GLM) were used to explore key predictors and underlying patterns influencing restoration willingness, rather than to produce generalized behavioral forecasts. As data-driven tools, these models are not designed to infer causality, and their performance depends heavily on the quality, structure, and balance of the input data. While the GBM model showed strong predictive accuracy and revealed non-linear relationships and feature importance, its findings should be interpreted as exploratory due to the cross-sectional nature of the dataset and the modest sample size.

Therefore, future studies using longitudinal data, expanded geographic coverage, and larger, more diverse samples are recommended to validate the findings and enhance predictive generalizability. Integrating mixed-method approaches could also provide deeper insight into the socio-cultural dimensions influencing restoration behavior. Moreover, future research should explore the long-term impact of restoration efforts on community perception, tracking changes over time to assess the effectiveness of awareness campaigns and policy interventions. Expanding the study across different geographic and socio-economic contexts would provide a broader understanding of coastal forest conservation challenges. Additionally, integrating remote sensing and GIS-based analyses with machine learning models could enhance the precision of ecosystem service valuation. Research should also investigate the role of traditional ecological knowledge in shaping conservation attitudes, as well as the potential for community-led restoration initiatives to improve sustainability outcomes. Finally, exploring the economic trade-offs between conservation and livelihood needs can help create balanced policies that ensure both ecological and socio-economic benefits.

5. Conclusions

This study underscores the importance of coastal forests in Bangladesh, providing essential ecosystem services such as provisioning, regulating, cultural, and supporting benefits. Understanding community perceptions of these services and their willingness to support restoration efforts is vital for sustainable conservation planning. Using a combination of focus group discussions, principal component analysis, and machine learning models, this study identified key factors influencing community perceptions and restoration willingness. A structured survey was conducted among coastal communities, comparing those living close to the forest with those farther away. The results revealed that proximity significantly shaped perceptions, with communities near the forest valuing provisioning services (e.g., food, fuelwood, and water) more highly, while those farther away prioritized timber. Regulating services, particularly protection from tidal surges and erosion control, were recognized as highly important by both groups. Cultural services, such as esthetics and tourism, were more appreciated by those living near the forest, whereas supporting services, especially bird habitats, were widely valued.

Socio-economic factors significantly influenced perception levels. Education showed a negative correlation, while age, dependence on forest resources, and awareness of restoration positively influenced willingness to support restoration. Machine learning analysis, particularly the Gradient Boosting Machine (GBM) model, outperformed other models in predicting restoration willingness, identifying distance from the forest, perceived importance of supporting services, and awareness of restoration as the top three predictors (accounting for 55.56% of variation). Policy recommendations include enhancing community-driven restoration efforts, integrating environmental education, and providing financial incentives to encourage participation. Targeted awareness programs should focus on younger and older demographics while addressing the concerns of middle-aged groups. By leveraging machine learning insights, this study bridges ecological conservation with socio-economic priorities, ensuring the long-term resilience and sustainability of Bangladesh’s coastal forests.