Comparative Analysis of Machine Learning Techniques for Fault Detection in Solar Panel Systems

Abstract

The utilization of Machine Learning (ML) classifiers offers a viable approach to improving diagnostic accuracy and system dependability in the pursuit of optimizing problem detection in solar panel systems. This work aims to conduct a thorough assessment of different Machine Learning (ML) classifiers in order to determine the most efficient models for detecting faults in solar panel systems. We rigorously tested and analyzed the classifiers AdaBoost, GaussianNB, Logistic Regression, Support Vector Classifier (SVC), Multi-Layer Perceptron (MLP), Decision Tree (DT), K-Nearest Neighbors (KNN), Random Forest (RF), and Extra Trees (ET). We evaluated the classifiers using their F1 scores, a crucial metric for measuring model performance in imbalanced class scenarios commonly encountered in fault detection tasks. The results show that the Decision Tree (DT), KNN, Random Forest (RF), and Extra Trees (ET) classifiers worked better than expected. All of them got perfect F1 scores of 1.000, which shows how well they can find bugs. On the other hand, AdaBoost demonstrated a lower F1 score of 0.591, suggesting possible constraints in its use for detecting faults in solar panel systems. This study advances fault detection in solar panels, enhancing system reliability and reducing maintenance costs. It also guides the development of sophisticated diagnostic tools, boosting solar technology adoption.