A machine learning framework for residential district cooling: Forecasting consumption, explaining drivers, and evaluating decarbonization pathways

District Cooling Systems (DCS) in the Middle East, while energy-efficient, are significant contributors to carbon emissions. This study introduces a novel framework to decarbonize DCS operations by integrating predictive machine learning, explainable AI (XAI), and renewable energy planning, all grounded in extensive real-world data. Leveraging a unique dataset from 59 residential buildings in the UAE—including energy consumption, climate variables, and building features—we developed a high-fidelity cooling load forecasting model. Following a rigorous chronological validation methodology, the Random Forest model was identified as the most robust, achieving a strong performance (R 2 = 0.8256, RMSE = 11,668.31). Outdoor temperature was confirmed as a primary driver of cooling load (r = 0.79). XAI analysis using SHAP confirmed that temperature, month, and occupancy rate are the most influential predictors. Crucially, scenario-based evaluations found that, under annual-energy assumptions without temporal matching, a 25 % solar offset yields greater carbon savings than an aggressive 30 % energy efficiency improvement. These findings demonstrate a scalable, data-driven blueprint for enhancing the operational efficiency and sustainability of residential DCS. The framework provides an interpretable, evidence-based tool for policymakers to support climate goals, such as the UAE Vision 2050, in arid urban regions.