An XAI-driven diagnostic framework to investigate the predictive power of building features to enhance EPC ratings in detached houses

Abstract

Energy performance in detached homes is critical for reducing carbon emissions in United Kingdom. However, understanding the complex factors that affect Energy Performance Certificate (EPC) ratings remains limited. Detached homes face unique challenges due to their larger floor areas and greater environmental exposure. Despite the significance of EPCs in driving energy efficiency, the diagnostic analysis of feature interactions at the class level (A–G) is underexplored, especially in detached homes. This study addresses this gap by employing predictive explainability to provide a detailed, rating class-wise diagnostic analysis of the predictive power of structural and operational features for detached homes. We investigate key factors such as CO2 emissions per floor area, heating costs, window, floor, walls efficiency, and construction age, and explore how these features interact to drive EPC ratings. Our findings show that CO2 emissions and heating costs are the primary drivers of EPC classification, but their impact varies across EPC bands. Detached homes in lower EPC categories (E–G) exhibit heightened sensitivity to high emissions and inefficient heating, while properties in EPC A and B benefit from improved insulation and efficient systems. This study introduces an innovative diagnostic framework that not only identifies key predictive features for each EPC class but also uncovers the synergistic effects of feature combinations. The results provide actionable insights for retrofit strategies and policy interventions, particularly for detached homes, offering a roadmap for improving energy efficiency and advancing sustainable energy practices.