AI-driven environmental performance index using deep autoencoders for monitoring sustainable development goals

This study develops an autoencoder-based environmental performance index (AE_EPI) to assess environmental performance across 38 countries in five regions. Using a deep autoencoder, AE_EPI captures complex nonlinear relationships in high-dimensional, multi-region data and is benchmarked against a principal component analysis-based index (PCA_EPI). Linear Regression, Random Forest (RF), and XGBoost, combined with SHAP analysis, are used to identify key environmental drivers. Results show that AE_EPI outperforms PCA_EPI in reconstruction accuracy, stability, and robustness to noise, while RF yields the most consistent predictions. Globally, electricity access and renewable energy are the dominant drivers. Regionally, governance and innovation are most important in East Asia; urbanization in South Asia; renewable energy in Central Asia; digitalization in MENA; and economic capacity, electricity access, and governance in Sub-Saharan Africa. These findings support region-specific, data-driven policies that advance energy transitions, strengthen institutional capacity, and address local environmental priorities for achieving the Sustainable Development Goals.