Integrating machine learning and sustainability indices for biomass energy forecasting in Türkiye: a data‐driven framework for agricultural, hydrological, and environmental policy design

Abstract Biomass energy is increasingly recognized as a strategic component of sustainable energy transitions and global efforts toward the Sustainable Development Goals (SDG‐7 and SDG‐13) and long‐term net‐zero emission targets. However, most national‐scale forecasting studies prioritize predictive accuracy while treating sustainability indicators as secondary inputs. To address this gap, this study proposes an integrated and explainable machine learning (ML) framework that combines biomass energy forecasting with sustainability assessment. The empirical analysis focuses on Türkiye during 2016–2024 using an annualized dataset derived from 108 monthly observations, including biomass energy capacity, agricultural productivity, water and waste‐to‐energy efficiency, greenhouse gas emissions, and macroeconomic indicators. Gradient boosting (GB), XGBoost, and CatBoost models were implemented to capture nonlinear interactions among energy, environmental, and agricultural variables that are difficult to represent using conventional statistical approaches. Model performance was evaluated using multi‐metric diagnostics and rolling time‐series validation. The GB model demonstrated robust out‐of‐sample performance with a rolling RMSE below 2 MW and consistently outperformed a naive trend benchmark. SHapley Additive exPlanations (SHAP) identified agricultural productivity and water and waste‐to‐energy efficiency as the dominant structural predictors, jointly accounting for more than 80% of total explanatory importance. Building on these explainability results, a novel Biomass Sustainability Index (BSI) was developed using empirically derived SHAP weights rather than subjective weighting schemes. The BSI exhibited a strong positive association with biomass energy capacity ( r = 0.87, R 2 = 0.81), while lagged analyses indicated that favorable sustainability conditions systematically precede capacity expansion. Uncertainty‐aware forecasting based on quantile GB generated probabilistic prediction intervals and future scenarios for 2025–2035. Results suggest continued biomass growth, with projected capacities ranging from approximately 1.8 GW under conservative assumptions to 2.14 GW under optimistic conditions. The proposed ML–BSI framework provides a transparent, interpretable, and transferable tool for sustainability‐oriented biomass energy planning by integrating prediction, explainability, and sustainability diagnostics within a unified analytical framework.