Accurate energy consumption forecasting is a central challenge in smart building design and cloud-based infrastructure management. This study proposes an interpretable and efficient regression model for Heating Load prediction using Shapley value–based feature selection. By modeling input features as cooperative players in a game-theoretic framework, the approach provides a principled measure of each variable’s contribution to the predictive task. Experiments conducted on the Energy Efficiency Dataset demonstrate that using only the top-ranked features, as determined by their Shapley values, preserves predictive performance while significantly reducing model complexity. The proposed model achieves high R \(^2\) and low error rates with improved interpretability, making it suitable for deployment in energy-aware AI systems.