Interpretable Data-Driven Models for Energy Performance Assessment in Residential Buildings

The assessment of buildings’ energy performance plays a critical role in achieving global sustainability goals, particularly in reducing carbon emissions and improving energy efficiency. In this context, various modelling approaches have been developed to evaluate building energy performance. Among them, data-driven models, such as machine learning (ML) algorithms, have gained significant attention in recent years due to their scalability, fast development process, and high predictive accuracy. However, a key limitation of these models is their limited interpretability, which can negatively affect their application particularly in decision-making and retrofit planning processes. To address this issue, SHapley Additive exPlanations (SHAP) has emerged as a promising approach for interpreting complex ML models by quantifying the contribution of each input feature to the model’s predictions. As a result, this study developed an XGBoost ML model that predicts energy performance of residential buildings in the UK with an R2 value of more than 0.98. After that, SHAP method was applied to explore and explain the effect of individual features on model outcomes, which highlighted that SHAP framework can be a strong complementary approach for enhancing the interpretability and practical applicability of black-box models in building energy performance analysis.