Sustainable Digital Business Ecosystems: Linking ESG Accounting, Data Analytics, and Value Creation Using Machine Learning Approaches

1.1. Research gap

Despite the growing body of literature on sustainability, digital transformation, and ESG integration, several critical gaps remain insufficiently addressed in current research.

First, prior studies largely examine ESG accounting, data analytics, and artificial intelligence (AI) as separate constructs, rather than as an integrated system within digital business ecosystems. For instance, ESG accounting has been widely discussed as a mechanism for measuring sustainability performance and corporate value (Renaldo, 2024; Comoli et al., 2023), while digital transformation and data analytics have been linked to improved ESG outcomes (Bindeeba et al., 2025; Agag et al., 2025). Similarly, AI-driven approaches have been recognized for enhancing ESG evaluation and sustainable decision-making (Rane et al., 2024; Suri & Sadriwala, 2025). However, limited studies have developed a holistic framework that simultaneously connects these three domains within a unified analytical model. Second, existing literature on digital transformation and sustainability tends to focus on linear relationships, overlooking the complex and nonlinear interactions among ESG practices, data analytics capabilities, and value creation. Empirical studies indicate that big data analytics capabilities exhibit nonlinear effects on ESG integration and sustainable innovation (Sun & Lim, 2026), suggesting that traditional statistical approaches may be insufficient. Although machine learning has been increasingly applied in sustainability contexts, such as ESG-based decision prediction (Hong et al., 2022), its application in modeling sustainable value creation within digital business ecosystems remains limited. Third, while recent research highlights the role of AI in enabling sustainable accounting and digital finance (J. Nair et al., 2025; Necula et al., 2025), there is still a lack of quantitative modeling frameworks that explicitly link AI-driven data analytics with ESG accounting and measurable value outcomes. Existing frameworks often remain conceptual (Mahwish et al., 2025) or focus on specific sectors such as fintech or manufacturing, thereby limiting their generalizability across broader digital ecosystems. Fourth, the concept of sustainable value creation itself remains underexplored in integrated digital contexts. Although studies have examined ESG performance and its economic implications (Kwilinski et al., 2023; Setiawan, 2026), few have incorporated multi-dimensional value creation models that combine financial, environmental, and social metrics within AI-driven analytical systems. Furthermore, emerging perspectives such as AI-enabled ecosystem innovation and triple helix collaboration suggest that sustainability outcomes are increasingly shaped by complex, data-driven interactions among multiple stakeholders (Barile et al., 2026; Mais et al., 2026), yet these dynamics are rarely captured in empirical models.

Therefore, this study addresses these gaps by developing a machine learning-based quantitative model that integrates ESG accounting, data analytics capabilities, and sustainable value creation within digital business ecosystems. By doing so, it moves beyond fragmented and linear approaches toward a holistic, nonlinear, and data-driven framework, contributing to both sustainability accounting and digital business research.

1.2 Research questions

Based on the identified gaps, this study is guided by the following research questions:

1.3. Hypothesis development

ESG accounting provides a structured framework for measuring sustainability performance and linking it to corporate value. Prior studies suggest that organizations with strong ESG practices tend to achieve better financial and non-financial outcomes (Renaldo, 2024; Kwilinski et al., 2023). Moreover, ESG-oriented accounting enhances transparency and stakeholder trust, which are critical drivers of value creation in digital ecosystems.

The implementation of ESG accounting requires advanced data processing, integration, and reporting systems. Digital accounting systems supported by AI and analytics enable organizations to manage complex ESG data more effectively (De Silva et al., 2025; Homotiuk & Mazuryk, 2025). As a result, firms with stronger ESG accounting practices are more likely to develop enhanced data analytics capabilities.

Data analytics capability enables firms to extract actionable insights from large datasets, improve decision-making, and optimize resource allocation. Studies have shown that data-driven approaches significantly enhance sustainability performance and business model innovation (Soni, 2025; Sun & Lim, 2026). In digital business ecosystems, analytics capabilities play a crucial role in translating ESG data into measurable value.

Data analytics capability serves as a critical mechanism through which ESG accounting influences value creation. ESG data alone may not generate value unless it is effectively analyzed and utilized. AI-driven analytics tools enhance the predictive and explanatory power of ESG information (Rane et al., 2024; Suri & Sadriwala, 2025), thereby strengthening its impact on value creation.

Traditional linear models may fail to capture the complexity of relationships in digital ecosystems. Machine learning approaches are capable of modeling nonlinear interactions and identifying hidden patterns in ESG and business data (Hong et al., 2022; Urbanovič & Holubčík, 2026). Therefore, the relationship between ESG accounting, data analytics, and value creation is expected to be nonlinear and better captured using machine learning techniques.

The integration of AI within digital business ecosystems enhances the alignment between ESG practices and value creation. AI-enabled systems facilitate real-time data processing, improve ESG measurement accuracy, and support sustainable decision-making (Nevi et al., 2025; Jebadurai & David, 2026). This integration strengthens the overall impact of ESG accounting and data analytics on sustainability outcomes.

2.1. Research design

This study adopts a quantitative, data-driven research design using machine learning techniques to model the relationships between ESG accounting, data analytics capability, and sustainable value creation in digital business ecosystems. Unlike conventional econometric approaches, machine learning enables the identification of nonlinear patterns and complex interactions among variables, which are increasingly relevant in sustainability and digital transformation contexts (Li, 2025; Lee et al., 2022). The research framework integrates ESG-related indicators, data analytics capability variables, and value creation metrics into a predictive modeling structure. This approach aligns with prior studies emphasizing the role of machine learning in ESG evaluation, sustainable investment decision-making, and predictive analytics (Patel et al., 2026; Twinamatsiko & Kumar, 2022).

2.2. Data and variable construction

The dataset consists of firm-level observations derived from ESG disclosures, financial reports, and digital capability indicators. Variables are operationalized as follows at Table 1 below:

2.3. Machine learning models

To capture nonlinear relationships and improve predictive accuracy, this study employs two main machine learning algorithms:

2.3.1. Random forest (RF)

Random Forest is an ensemble learning method that constructs multiple decision trees and aggregates their predictions to improve accuracy and reduce overfitting (Siswoyo et al., 2025).

The prediction function can be expressed as:

where:

Random Forest is particularly effective for handling high-dimensional ESG datasets and identifying variable importance, which is critical in sustainability analytics (Słoniec et al., 2025).

2.3.2. Extreme gradient boosting (XGBoost)

XGBoost is a gradient boosting algorithm that builds models sequentially, where each new model minimizes the errors of previous models (Chen & Guestrin, 2016; adopted in ESG research by Lee et al., 2022).

The objective function is defined as:

where:

XGBoost is highly efficient in modeling nonlinear ESG relationships and interaction effects, making it suitable for sustainability research (Li, 2025; Patel et al., 2026).

2.4. Model evaluation

To assess model performance, the following evaluation metrics are used:

These metrics provide a comprehensive evaluation of predictive accuracy and model robustness (Gomes et al., 2025).

2.5. Feature importance and explainability

To enhance interpretability, this study employs feature importance analysis to identify key drivers of sustainable value creation. In addition, explainable AI techniques such as SHAP (Shapley Additive Explanations) are used to interpret the contribution of each variable in the prediction model. This approach is consistent with prior research emphasizing the importance of explainable machine learning in ESG and sustainability analytics (Davidescu et al., 2025).

Figure 2 presents the ESG data lifecycle and analytics process within digital business ecosystems, illustrating how data flows from capture and storage to reporting, rating, and sustainable investment decisions. The framework highlights key stages, including data sourcing, processing, quality control, assurance, and disclosure, supported by data analytics and machine learning techniques. This lifecycle demonstrates how ESG data is transformed into actionable insights through analytical processes, enabling firms to optimize decision-making and sustainability performance. The integration of analytics and reporting mechanisms reflects the increasing importance of data-driven ESG management and digital transformation (Ertz et al., 2025; Petcu et al., 2024). Furthermore, the model aligns with research emphasizing the role of machine learning in ESG evaluation and sustainable finance, where data pipelines and analytics capabilities are critical for generating predictive insights and supporting investment decisions (Jaiswal et al., 2025; Li, 2025).

Figure 2. ESG data analytics lifecycle and machine learning integration framework.

2.6. Research procedure

Figure 3 illustrates the structured research procedure adopted in this study, encompassing five sequential stages: data collection and preprocessing, feature engineering, model training, model validation, and interpretation. The process begins with ESG data normalization and missing value handling to ensure data quality and consistency. Subsequently, feature engineering transforms raw ESG indicators into composite indices and standardized variables suitable for machine learning models. The model training stage employs ensemble learning techniques, namely Random Forest and XGBoost, to capture complex and nonlinear relationships within ESG data. Model validation is conducted using k-fold cross-validation and performance comparison based on established evaluation metrics, including MAE, RMSE, and R². Finally, the interpretation stage utilizes feature importance and SHAP analysis to enhance model transparency and explainability. This framework reflects a data-driven and AI-enabled approach to sustainability analytics, aligning with prior studies emphasizing the importance of integrating machine learning into ESG analysis and decision-making processes (Lee et al., 2022; Li, 2025; Patel et al., 2026). It also highlights the critical role of explainable AI in improving the interpretability and reliability of sustainability models (Davidescu et al., 2025).

Figure 3. Research procedure for esg data analytics using machine learning models.

3.1. Descriptive statistics and correlation analysis

As shown in Table 2, ESG accounting exhibits a strong positive correlation with sustainable value creation (r = 0.68) and data analytics capability (r = 0.62), providing initial support for H1 and H2. Furthermore, data analytics capability shows the strongest correlation with value creation (r = 0.71), reinforcing its mediating role (H4). These findings are consistent with prior studies emphasizing the role of ESG and digital capabilities in enhancing firm performance (Kwilinski et al., 2023; Pesqueira & Sousa, 2024).

3.2. Model performance evaluation (based on equations MAE, RMSE, R²)

Based on the evaluation metrics defined in Equations (MAE, RMSE, and R²) in the Methods section, Table 3 shows that XGBoost achieves the lowest prediction error (MAE = 3.95; RMSE = 5.21) and the highest explanatory power (R² = 0.87). This confirms that XGBoost more effectively minimizes the loss function, compared to Random Forest and linear regression. The results validate H5, indicating that machine learning models capture nonlinear relationships more effectively than traditional models (Lee et al., 2022; Li, 2025).

3.3. Feature importance and contribution analysis

As presented in Table 4, ESG Governance contributes the highest gain (26.3%), indicating its dominant role in reducing prediction error within the XGBoost model. Data analytics capability also shows substantial importance (22.1%), confirming its role as a key mediator (H4). The results support findings from Gomes et al. (2025) and Saxena et al. (2022), emphasizing governance and analytics as central drivers of sustainability performance.

3.4. Mediation and interaction effects (model-based interpretation)

Table 5 demonstrates that ESG accounting has both direct (0.42) and indirect effects (0.29) on sustainable value creation through data analytics capability. The total effect (0.71) confirms a strong mediation mechanism, supporting H4. This result aligns with prior research suggesting that ESG data must be processed through analytics to generate value (Soni, 2025; Patel et al., 2026).

3.5. Nonlinear relationship analysis (partial dependence insights)

As shown in Table 6, the relationship between ESG and value creation follows a sigmoid pattern, indicating that ESG initiatives only produce substantial value after reaching a critical threshold (~60). Similarly, data analytics capability exhibits increasing returns, confirming nonlinear dynamics as hypothesized in H5. These findings are consistent with Sun & Lim (2026) and Li (2025).

3.6. Random forest aggregation mechanism (consistency check)

Table 7 shows that increasing the number of trees improves model stability and accuracy. However, the marginal improvement decreases beyond 100 trees, indicating convergence. This supports the robustness of ensemble learning in ESG prediction (Siswoyo et al., 2025).

3.7. Robustness and model comparison

As shown in Table 8, XGBoost achieves the best balance between predictive accuracy and nonlinear modeling capability, confirming its suitability for ESG and sustainability analysis. This supports H5 and H6, demonstrating that AI-driven models enhance analytical performance (Davidescu et al., 2025; Patel et al., 2026).

Theoretical contribution

This study makes three key theoretical contributions:

Practical implications

From a managerial perspective, the findings suggest that firms should: Invest in data analytics and AI capabilities to maximize the value of ESG initiatives. Focus on governance quality as a key driver of sustainability performance. Recognize that ESG benefits emerge after reaching a critical threshold, requiring long-term commitment.

Revisiting the research gap

This study successfully addresses the gaps identified earlier by: Integrating ESG accounting, data analytics, and AI into a single model. Demonstrating nonlinear relationships using machine learning. Providing empirical evidence on sustainable value creation in digital ecosystems Thus, it moves beyond prior conceptual and fragmented approaches toward a holistic, data-driven, and predictive framework for sustainability research.

Policy implications

The findings of this study offer several important implications for policymakers, regulators, and business leaders in promoting sustainability within digital economies. First, policymakers should encourage the adoption of standardized ESG accounting frameworks that are compatible with digital reporting systems. Integrating ESG metrics into digital infrastructures will enhance transparency, comparability, and accountability across firms. Second, governments and regulatory bodies need to support the development of data analytics and AI capabilities, particularly among small and medium-sized enterprises (SMEs). Public investment in digital infrastructure, data ecosystems, and AI literacy can significantly improve firms’ ability to leverage ESG data for sustainable decision-making. Third, regulatory frameworks should promote the use of AI and machine learning in sustainability assessment, while ensuring ethical standards, data governance, and algorithmic transparency. This is essential to balance innovation with accountability in AI-driven sustainability practices. Fourth, the identification of ESG threshold effects suggests that sustainability policies should adopt a long-term perspective, providing incentives for firms to reach critical levels of ESG performance rather than focusing solely on short-term compliance.

Cross sector collaboration particularly through triple helix models (government–industry–academia) should be strengthened to accelerate innovation in sustainable digital ecosystems and foster data-driven value creation.

Limitations and future research

Despite its contributions, this study has several limitations that provide opportunities for future research. First, the study relies on secondary ESG and firm-level data, which may be subject to reporting bias and differences in ESG measurement standards. Future studies could incorporate primary data or alternative ESG scoring methods to enhance robustness. Second, while this study employs advanced machine learning models, it focuses primarily on tree-based algorithms (Random Forest and XGBoost). Future research could explore deep learning models, such as neural networks, to capture more complex patterns in sustainability data. Third, the analysis is conducted at the firm level and does not fully capture ecosystem-level dynamics, such as interactions among multiple stakeholders within digital platforms. Future studies could adopt network analysis or system dynamics approaches to better understand these interactions. Fourth, although this study identifies nonlinear relationships and threshold effects, it does not explicitly examine causal mechanisms. Future research could integrate causal inference techniques, such as quasi-experimental designs or hybrid ML-econometric models.

Future research could extend this framework to specific sectors such as fintech, manufacturing, or platform economies or conduct cross-country comparative studies to explore institutional and contextual differences in ESG-driven value creation.