Evaluating Strategic Retrofit Measures for Energy-Efficient Residential Buildings with Artificial Intelligence

The global building sector is one of the main contributors to annual global greenhouse gas emissions, yet homeowners remain hesitant regarding specific retrofit measures to reduce carbon emissions. This is unsurprising as the link between retrofits that reduce energy consumption and corresponding economic and ecological benefits remains elusive. Therefore, this study addresses the intersection of building energy performance, carbon emission reduction, and financial subsidies by quantifying expected energy savings based on specific energy-related retrofits with a real-world dataset containing 25,000 German residential buildings. The simulated energy savings for specific retrofit measures are based on a novel feature value substitution methodology and three sophisticated machine learning models, namely XGBoost, CatBoost, and LightGBM. This study then combines potential ecological gains, household investment budgets, and expected local governmental subsidies into a single informative yet comprehensible retrofit index to overcome the uncertainty regarding retrofits. The results show that glazing is the most impactful feature for potential energy savings of residential buildings, followed by heating system changes from oil to electric heating pumps. In contrast to the neglectable impact of better facade conditions on building energy performance, roof and wall insulation improvements lead to significantly lower energy consumption. This study underscores potential ecological savings of targeted retrofit measures and enables practitioners to cut expenses and reduce the associated financial risks.