Comparative Analysis of Quadratic and linear Discriminant Analysis in Power line Fault Detection

Precise fault detection system is crucial in safeguarding the reliability of electrical transmission networks. Machine learning and other Artificial intelligence techniques have played a central role in detecting faults in electric power lines. However, Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) remains underexplored in power line fault detection community. In view of this, the study presents a fault detection approach involving utilization of mutual information for feature selection followed by the LDA and QDA for binary classification of historical fault patterns of Transmission Company of Nigeria (TCN) obtained from National Control Center (NCC), Abuja, Nigeria. Mutual information was used to score the relevancy of each features before reference threshold was set to reduce the features from ten to four thereby enhancing the performance of the classifiers used. The result showed that QDA performed well in binary classification of power line fault while LDA offered a lackluster performance. Based on this, the deployment of mutual information, a non-parametric feature selection technique significantly enhanced the performance of QDA, and to a small extent, the performance of LDA. The findings in this study provides insight on the potentials of multivariate Gaussian based algorithms in power line fault detection assignment.