The proven superiority of DC microgrids over AC microgrids, attributed to their energy efficiency, flexibility, simplicity, reduced complexity, high power transfer capability, and freedom from voltage, frequency regulation, and synchronization problems, contrasts with the significant challenge of protecting DC microgrids. Traditional protective devices (PDs) exhibit low sensitivity and selectivity when dealing with high-impedance faults, PV array string-to-string faults (SS faults), and string-to-ground faults (SG faults). The stochastic nature of distributed energy resources (DERs) further complicates the differentiation between PV array string faults and high-impedance distribution line faults. In PV systems, protection mechanisms like fuses and residual current detectors are available for detecting substantial fault currents. However, these mechanisms fail to recognize faults when solar and/or fault mismatches are small and fault resistance is high. Consequently, traditional protection systems prove inadequate in detecting problems during cloudy and low-irradiance conditions, posing a risk of photovoltaic fires and compromising dependability. To address this challenge, we propose a fault detection scheme for PV systems that involves feature extraction using suitable discrete wavelet techniques, capturing distinctive attributes from voltage and current profiles. Additionally, the scheme employs appropriate ensemble machine learning methods for classifying various faults in both the PV system and distribution line.