Explainable AI-Based Predictive Maintenance for Public EV Charging Infrastructure Using Real OCPP Data with SHAP Analysis

Abstract

Crucially impacting user satisfaction and the downtime of electric vehicle (EV) charging stations, the reliability of EV infrastructure inevitably drives the transition toward car- bon-free emission vehicles. Proposing an explainable artificial intelligence (AI)-based predictive maintenance frame- work for public EV charging infrastructure using real-world Open Charge Point Protocol (OCPP) data with SHAP analysis, this thesis combines large-scale feature engineering, multiple machine learning models, SHAP-based feature selection, and decision-thresh- old optimization to predict station-level failures in advance and identify the key factors driving these failures; it is evaluated using real data from ENGIE Vianeo, considering both an initial dataset of 104 charging stations and an extended dataset of 291 stations under different observation periods. Experimental results demonstrate that tree-based ensem- ble methods outperform linear and distance-based models, with Random Forest achiev- ing the best performance (0.800 F1-score and 0.828 recall). The integration of SHAP- based feature selection and threshold optimization consistently improves model perfor- mance, particularly on the larger dataset. From an operational perspective, the results highlight the potential economic impact of predictive maintenance. Based on observed downtime of approximately 32,000 days per year and an estimated annual cost of €243,000, the proposed framework could enable the early detection of up to 82.8% of faults. Assuming that 30% to 50% of predicted failures can be prevented, this corresponds to potential annual savings between €60,000 and €100,000. Overall, the proposed framework demonstrates that explainable machine learning techniques can effectively support predictive maintenance in EV charging infra- structure, offering both improved reliability and significant operational benefits. Keywords: Predictive Maintenance, Electric Vehicle (EV) Charging Infrastructure, Ex- plainable Artificial Intelligence (XAI), SHAP Analysis. Islam Elmorshidy 02/06/2026