Residential cooling demand in southern European cities is projected to rise sharply under climate change, yet building-level evidence to guide municipal adaptation remains scarce. This thesis develops an explainable machine-learning framework that projects residential cooling demand for Milan's EPC-labelled residential building stock under CMIP6 climate scenarios and translates the projections into planning-scale decision support. The framework integrates 342,684 CENED+2 energy performance certificates, reduced through building-footprint matching to 19,063 EPC-labelled buildings within the 53,041-polygon DBT2012 inventory, with ERA5-Land reanalysis (1990–2024) and a twelve-model NEX-GDDP-CMIP6 ensemble under SSP2-4.5 and SSP5-8.5. A two-phase design separates extent from intensity. Phase 1 trains a tuned XGBoost classifier for cooling-presence estimation, validated across the EPC-labelled corpus through held-out, out-of-fold, and spatial cross-validation and reaching a test ROC-AUC of 0.972 and a spatial cross-validation ROC-AUC of 0.969.