Multiclass Electrical Fault Classification in Three-Phase Power Systems Using Random Forest, Support Vector Machine, and XGBoost

Electrical fault classification in three-phase power systems remains challenging because different fault conditions can exhibit similar current and voltage characteristics, limiting the diagnostic capability of conventional binary fault detection. This study aimed to compare the performance of Random Forest, Support Vector Machine, and Extreme Gradient Boosting (XGBoost) for multiclass electrical fault classification and to identify the most effective and stable model. The publicly available Electrical Fault Detection and Classification dataset obtained from the Kaggle repository was used in this study. The dataset contains 7,861 observations with six electrical features comprising three-phase currents (Ia, Ib, and Ic) and voltages (Va, Vb, and Vc), which were used to classify six operating conditions: Normal, Line-to-Ground, Line-to-Line, Double Line-to-Ground, Three-Phase, and Three-Phase-to-Ground faults. The models were evaluated using accuracy, precision, recall, F1-score, and five-fold stratified cross-validation, complemented by confusion-matrix and feature-importance analyses. Random Forest achieved the best test performance with an accuracy of 86.78% and an F1-score of 86.76%, while cross-validation produced a mean accuracy of 87.55% and a mean F1-score of 87.53%. Class-level analysis revealed that Three-Phase and Three-Phase-to-Ground faults were the most difficult conditions to distinguish because of substantial overlap in their electrical characteristics. The findings demonstrated that Random Forest provided the most effective and stable classification performance and that combining performance, stability, and class-level analyses provided deeper insight into multiclass electrical fault diagnosis.