Identifying residential building energy retrofit priorities across climate zones and end uses using machine learning and explainable AI

Energy retrofits of existing residential buildings are essential for reducing energy consumption and greenhouse gas emissions. However, limited resources require effective prioritization of retrofit measures. As retrofit effectiveness varies by climate zones, end uses, building characteristics, and future climate conditions, uniform retrofit strategies are often inefficient at national scales. Unlike previous studies that typically focus on a single climate condition or energy end use, this study develops a data-driven framework systematically comparing climate-zone and end-use-specific retrofit priorities under both historical and future climate scenarios, enabling climate-responsive retrofit planning and policy development. A synthetic database of over 72,000 parametric simulations of residential building archetypes is used to train climate-zone-specific machine learning (ML) models for predicting heating and cooling energy use intensity (EUI). Explainable artificial intelligence, based on SHapley Additive exPlanations (SHAP), is applied to interpret model predictions and quantify parameter importance and directional influence. Results show substantial climate-zone-dependent differences in retrofit priorities. For cooling EUI, window SHGC and south-facing façade window-to-wall ratio (FDWR) are the two most influential features across all climate zones, exhibiting SHAP distributions approximately 1.5–2 times wider than those of other features. Cooling system efficiency and foundation type are additional key retrofit priorities in warmer and transitional climates (CZ 4–6), whereas ventilation rate plays a larger role in colder zones (CZ 7 A–8). For heating EUI, window U-value dominates in CZ 4–5, with SHAP ranges approximately 2–3 times wider than other variables, while foundation type becomes the primary driver in CZ 6–8. Under future climate scenarios, mechanical cooling efficiency gains importance in CZ 4-7 A, while heating performance in colder zones remains dominated by envelope, foundation, and ventilation system with heat recovery. Overall, the proposed ML-XAI framework provides transparent insights to support regionally differentiated climate-responsive retrofit strategies and evidence-based energy policy development.