Predicting Thermal Energy Demands for Residential Buildings in Egypt: A Machine Learning Perspective

Buildings account for a significant share of global energy use, with heating and cooling loads representing the most significant contributors to residential energy demand. Accurate prediction of these loads is therefore essential for designing efficient HVAC systems and supporting energy-saving strategies. Traditional estimation methods, however, rely on mathematical models or empirical correlations that often fail to capture the complex interactions among building design variables, occupancy patterns, and environmental conditions. This study addresses this gap by conducting a comprehensive experimental assessment of multiple ML algorithms for predicting residential heating and cooling loads. A simulated dataset was generated from diverse building configurations, assuming a residential setting located in Cairo, Egypt, varying in geometry, orientation, insulation, and glazing parameters. Data preprocessing, normalization, and systematic feature selection were applied to identify the most relevant predictors. Nine supervised learning methods, including ensemble, kernel-based, and linear models, were trained and evaluated using rigorous cross-validation and multiple performance metrics. The findings indicate that ensemble-based approaches offer superior predictive performance, while feature ranking highlights the strong influence of window-to-wall ratio and glazing distribution. These results emphasize the role of data-driven methods in enhancing energy-efficient building design and provide actionable insights for practitioners in architecture, engineering, and energy management.