<p>Fault detection and accurate localization in photovoltaic (PV) systems remain challenging, especially when the irradiance is low and during Maximum Power Point Tracking (MPPT) operation, when several electrical faults remain concealed. Among these, line-to-line (LL) and line-to-ground (LG) faults are especially critical as they may escape conventional protection schemes at low fault currents, leading to power losses, reduced system efficiency, and potential fire hazards. To address these challenges, this paper intends to offer a solution for detection and identification of string level position of LL and LG faults using supervised classification learners. The proposed approach utilizes system-level measurements, including irradiance, temperature, string currents, DC voltage, and DC current, for dataset generation. Extensive simulations are conducted by incorporating varying weather conditions, fault resistance levels, and string degradation scenarios. Multiple classification learners are evaluated for fault identification, followed by dedicated localization models for faulty string detection. Among the tested models, the neural network demonstrates superior performance, achieving a fault classification accuracy of 93.98%. Furthermore, the proposed localization framework attains accuracies of 98.31% for intra-string LL faults, 98.34% for cross-string LL faults, and 96.74% for LG faults. The robustness of the proposed method is further validated under conditions of missing data, sensor failures, communication delays, and varying training–testing data splits, in which the model maintains consistent performance. The results confirm that the proposed data-driven framework provides an effective and robust solution for accurate fault detection and localization in PV systems operating under practical field conditions.</p>