ML-driven energy prediction model for climate-based affordable passive envelope design in hot regions

Abstract

• ML model predicts cooling loads using envelope and climate-specific features. • Gradient boosting achieves R 2 = 0.98 with 35 % error reduction. • Climate data improves prediction accuracy by up to 20%. • Orientation-based U-value optimization cuts cooling loads by 25%. • Strategy reduces operational CO 2 by 19% and embodied CO 2 by 37%. As an alternative to traditional numerical and simulation-based methods, Machine Learning (ML) emerges as a powerful AI-driven approach for optimizing energy-efficient building design. Although the building envelope plays a critical role in controlling thermal behavior, ML applications that explicitly uncover the correlation between envelope-specific parameters and cooling energy demand remain limited. This study addresses this gap by proposing a methodological ML-driven energy prediction framework for optimizing passive building envelope strategies tailored to hot-desert contexts, using Egypt as a representative case. By examining the combined effects of thermal transmittance (U-value) and façade orientation, the model evaluates their influence on cooling energy demand while capturing climatic variability across eight hot-desert climate zones. To ensure robust predictions, physics-based and experimentally validated EnergyPlus simulations are integrated with advanced ML techniques, bridging the gap between computational modeling and data-driven prediction. Four algorithms are tested, with gradient boosting achieving the highest performance (R 2 = 0.99) and reducing prediction error by up to 35 %. Excluding climate data lowers accuracy by approximately 20 %, underscoring its importance in ML-based energy modeling even within the same hot climate category. The framework employs orientation-specific envelope configurations that reduce cooling loads by up to 25 %, operational CO 2 emissions by 15–19 %, embodied CO 2 by 29–37 %, and annual costs by approximately $532. This approach supports cost-effective, climate-adaptive passive envelope strategies that enhance performance without excessive material use. The framework can enable more informed early-stage design decisions and advance ML-integrated sustainable building practices by providing a rapid yet accurate alternative to conventional simulations.