This paper explores the detection and classification of faults in photovoltaic (PV) arrays using machine learning classifiers, focusing on Support Vector Machines (SVMs) and Extreme Gradient Boosting (XGBoost). The study addresses the challenge of accurately detecting faults on the Direct Current (DC) side of PV systems, which is often overlooked in previous research. To enhance the performance of these classifiers, the authors incorporate optimization algorithms such as the Bees Algorithm (BA) and Particle Swarm Optimization (PSO) to fine-tune the hyperparameters of SVMs and XGBoost models. The research involves building a small-scale PV array to simulate real-world complexities and introduce specific faults, generating a realistic dataset for training the machine learning algorithms. The proposed methods are evaluated using various performance metrics, including accuracy, sensitivity, and specificity, to assess their effectiveness in fault detection and classification. The results demonstrate that the Bees Algorithm significantly improves the accuracy of SVM and XGBoost classifiers, outperforming the PSO algorithm in this context. The study highlights the potential of hybrid methods combining multiple algorithms to address the challenges of fault detection in PV systems, particularly in distinguishing complex faults with similar I-V curves.This paper explores the detection and classification of faults in photovoltaic (PV) arrays using machine learning classifiers, focusing on Support Vector Machines (SVMs) and Extreme Gradient Boosting (XGBoost). The study addresses the challenge of accurately detecting faults on the Direct Current (DC) side of PV systems, which is often overlooked in previous research. To enhance the performance of these classifiers, the authors incorporate optimization algorithms such as the Bees Algorithm (BA) and Particle Swarm Optimization (PSO) to fine-tune the hyperparameters of SVMs and XGBoost models. The research involves building a small-scale PV array to simulate real-world complexities and introduce specific faults, generating a realistic dataset for training the machine learning algorithms. The proposed methods are evaluated using various performance metrics, including accuracy, sensitivity, and specificity, to assess their effectiveness in fault detection and classification. The results demonstrate that the Bees Algorithm significantly improves the accuracy of SVM and XGBoost classifiers, outperforming the PSO algorithm in this context. The study highlights the potential of hybrid methods combining multiple algorithms to address the challenges of fault detection in PV systems, particularly in distinguishing complex faults with similar I-V curves.