Comparative Analysis of Machine Learning Techniques for Binary Classification of Power Line Fault

In order to enhance the reliability of electric power distribution systems, this study aims to assess the performance of machine learning approaches for the classification of power line faults. Machine learning is an attractive candidate for efficient classification of faults in power lines. However, there is lack of clarity on which machine learning approach is the most suitable for power line fault detection assignment. The paper evaluates and compares the performances of Quadratic Discriminant Analysis (QDA), Linear Discriminant Analysis (LDA), and K-Nearest Neighbors (KNN) in differentiating fault and no-fault conditions in power line transmission systems. The transmission line fault data obtained from the National Control Center (NCC), Abuja, Nigeria, was normalized to guarantee consistency across features. Simultaneously, avoidance of the curse of dimensionality and enhancement of the performances of the models were ensured in this study through a reduction of the number of features from ten to four using mutual information and a reference threshold. Based on the performance metrics used, QDA possessed the best performance, followed closely by KNN. The work offers valuable insights into suitability of machine learning models for fault classification of power lines.