Sustainability assessment and ML-based prediction of building green roofs for mitigating subtropical urban heat

Buildings account for nearly 40% of global energy demand, with 20–40% attributed to space heating and cooling, making roofs a critical target for improving thermal performance. This study proposes an integrated sustainability assessment and machine learning (ML) framework to evaluate and predict green roof performance under subtropical climatic conditions. Experimental data on energy storage in roof slabs, exergy variation during space heating and cooling, and daily electricity cost savings were analyzed using combined energy–exergy–economic indicators. ML models, including Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Networks (ANN), achieved high predictive accuracy (R² > 0.95) with prediction errors below 5%. Global sensitivity analysis identified space cooling electricity savings as the dominant factor, contributing approximately 49% of total sensitivity, followed by exergy performance (31%). SHAP-based explainability showed that energy storage parameters governed ML predictions, accounting for up to 69% of total SHAP contribution. Experimental results demonstrated that macroencapsulated PCM-integrated roofs reduced indoor temperatures by up to 7.3°C and mitigated heat gain by 51.35%, with a cost recovery period of 5.7 years. The proposed framework offers a robust, interpretable, and computationally efficient decision-support tool for climate-resilient green roof design.