@inproceedings{3075b62b2ee7420483d219bd98685a57,
title = "Improved Power Transformer Fault Diagnosis Through Optimized Feature Selection in DGA Using SVM and KNN Techniques",
abstract = "This study enhances power transformer fault diagnosis by applying optimized feature selection in Dissolved Gas Analysis (DGA) with Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) algorithms. We evaluated these models across four distinct transformers for accurate identification of arcing, partial discharge (PD), thermal faults and normal condition. Both SVM and KNN achieved high overall accuracy (up to 96.54\% and 96.66\% respectively). Notably both algorithms showed exceptional precision and recall for “arcing” faults (especially in transformer 1 and 3), and improved consistency for “normal” and “arcing” classes in Transformer 2 and 4, highlighting the benefits of optimized feature selection. While “PD” and “thermal' faults were more challenging, these findings underline the effectiveness of DGA-based machine learning for transformer fault classification, indicating areas for further refinement.",
keywords = "Dissolved Gas Analysis, Fault Diagnosis, K-Nearest Neighbors, Power Transformer, Support Vector Machine",
author = "Malesa, \{Given Sipho\} and Lutendo Muremi and Vikash Rameshar",
note = "Publisher Copyright: {\textcopyright} 2025 IEEE.; 2025 IEEE AFRICON, AFRICON 2025 ; Conference date: 10-12-2025 Through 12-12-2025",
year = "2025",
doi = "10.1109/AFRICON66545.2025.11533749",
language = "English",
series = "IEEE AFRICON Conference",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
booktitle = "2025 IEEE AFRICON, AFRICON 2025",
address = "United States",
}