A hybrid approach for early fault prediction based on resistance behavior and machine learning

Fault detection in electrical power systems at an early stage plays a vital role in ensuring high operational efficiency, reliability, and cost-effectiveness by preventing unexpected system breakdowns. However, traditional fault diagnosis techniques based solely on raw voltage and current measurements may be sensitive to measurement noise and variations in operating conditions. To address this issue, this paper proposes a resistance-based statistical feature engineering framework integrated with machine learning for electrical fault classification. Phase-wise resistances derived from voltage and current measurements are used to extract statistical features including mean, maximum, minimum, standard deviation, and range. The extracted features are evaluated using six machine learning classifiers: Random Forest (RF), Extreme Gradient Boosting (XGBoost), CatBoost, Support Vector Machine (SVM), Logistic Regression (LR), and K-Nearest Neighbors (KNN). The proposed framework is evaluated on two datasets: the Electrical Fault Detection and Classification dataset from Kaggle and the Hybrid Electrical Fault Detection Dataset (Real + Synthetic) from Mendeley Data. Using an 80:20 training-testing split, Random Forest achieved the highest accuracy of 97.75% on the Kaggle dataset and 95.82% on the Hybrid Electrical Fault Detection Dataset. The results demonstrate that resistance-derived statistical features provide an interpretable and computationally efficient representation for reliable electrical fault diagnosis across different datasets. The proposed framework further demonstrates its potential for predictive maintenance and intelligent electrical fault monitoring.