An integrated machine learning framework for EV charging management

As the shift to electric mobility intensifies, unpredictable EV charging challenges grid stability. This study proposes a multi-layered machine learning framework balancing grid optimization and user service. First, session-level prediction models estimated energy and cost; XGBoost achieved the highest energy accuracy ([Formula: see text]), while Random Forest best predicted cost ([Formula: see text]). Second, a station-level forecasting model using XGBoost demonstrated exceptional precision for daily demand ([Formula: see text], MAE=0.90 kWh). Finally, K-Means clustering segmented drivers, revealing a user base dominated by Heavy Energy Users (43.5%) and Occasional Visitors (38.8%). This segmentation enables Charge Point Operators to design personalized services and demand response strategies. Overall, the framework integrates prediction, forecasting, and behavioral segmentation to support scalable, data-driven decisions. Ultimately, these insights equip utility providers and operators with the necessary tools to proactively manage load congestion and optimize capital expenditure planning.