Integration of Explainable AI (XAI) Models for ESG Reporting within a Sustainable Enterprise Information System Governance Framework

Abstract

Climate change demands transparent and evidence-based sustainability governance. This study develops an Explainable Artificial Intelligence–Enabled Decision Support System (XAI-EDSS) to predict and explain Indonesia’s carbon emissions within the Net Zero Emission 2060 framework. The system integrates Machine Learning (ML), Explainable AI (XAI), and Environmental, Social, and Governance (ESG) frameworks into a unified predictive–analytic platform that supports data-driven decision-making. Secondary data were obtained from Our World in Data (OWID), the World Bank, and the International Energy Agency (IEA) for the period 2000–2022, covering CO₂ emissions, Gross Domestic Product (GDP), and primary energy consumption. Several ML algorithms—Linear Regression, Random Forest, XGBoost, and Light Gradient Boosting Machine (LightGBM)—were employed, with hyperparameter optimization via Optuna and interpretability analysis using SHapley Additive exPlanations (SHAP). The results reveal that LightGBM achieved the best predictive performance with R² = 0.787 on the test set and CV Mean R² = 0.956, indicating high accuracy and strong  model stability. The SHAP-based Explainable AI analysis shows that energy consumption is the dominant driver of CO₂ emissions, while GDP contributes positively but with lower magnitude. The final model was deployed into an interactive Streamlit-based dashboard, integrating model outputs, forecasting scenarios (2025–2035: Business-as-Usual, Rapid Transition, and Global Recession), and ESG metric evaluation aligned with GRI 302, GRI 305, and TCFD frameworks. The implementation of XAI-EDSS enhances the transparency, interpretability, and accountability of ESG analytics, positioning it as a Responsible AI tool for sustainable policy reporting in Indonesia.