Estimating heterogeneous treatment effects of building energy efficiency retrofits using machine learning

Buildings are a major contributor to energy consumption and greenhouse gas emissions, with residential heating alone accounting for over 80% of Switzerland’s residential energy demand. Improving building efficiency through retrofitting is critical for achieving national climate goals. This study presents a machine learning-based framework to estimate the impact of energy efficiency measures on heat consumption. By integrating predictive models with causal inference techniques, we assess the heterogeneous treatment effects of various retrofitting actions, such as roof and facade insulation, and window replacement. Our framework combines domain-specific machine learning models for heat consumption estimation with causal inference methodologies to quantify the effectiveness of different retrofitting measures. We evaluate the approach using real-world heat consumption data from three Swiss regions, incorporating structural building attributes and retrofitting records. The results demonstrate the feasibility of this data-driven monitoring system while also highlighting current data limitations that hinder large-scale implementation. The ability to continuously track the actual energy savings from implemented retrofitting measures is essential for policymakers, building owners, and energy planners. This framework provides a scalable solution to bridge the gap between theoretical energy savings estimates and actual post-retrofitting performance. By improving the accuracy of energy efficiency evaluations, this approach supports more effective policy design, subsidy allocation, and progress toward energy and climate goals.