Feature Importance-Based Machine Learning Classification of Distribution Trans-former Health Conditions Using Oil Test Data
International Journal of Advanced Research in Science and Technology · 2026-07 · 15 (7) · 2169-2173
DOI: 10.62226/ijarst20262730 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Advanced Research in Science and Technology, 2026-07.
Abstract (Originalsprache)
Transformer health condition assessment is essential for improving the reliability of electric power distribution networks. This paper presents a feature-importance-based machine learning framework for classifying the health condition of oil-immersed distribution transformers using routine oil-test data. The dataset contained 3,373 cleaned diagnostic records after removing exact duplicates and incomplete key identifiers. Thirteen diag-nostic inputs were used, including dissolved gas analysis variables, dielec-tric breakdown strength, moisture, neutralization value, interfacial tension, color and 2-furfural. The continuous transformer health index was con-verted into five practical classes: Good, Acceptable, Need Caution, Poor and Critical. Seven classifiers were compared using an 80:20 stratified train-test split, namely logistic regression, k-nearest neighbours, support vector machine, multilayer perceptron, random forest, extra trees and gra-dient boosting. The best model was Gradient Boosting, which achieved 85.04% accuracy and 0.851 weighted F1-score on the independent test set. Feature importance analysis indicated that 2-furfural was the most influen-tial diagnostic variable, followed by dissolved gas indicators such as C2H2, C2H4, C2H6 and CH4, as well as oil-quality variables including dielectric breakdown strength, interfacial tension, neutralization value and moisture. The results show that machine learning can provide a practical decision-support layer for rapid transformer health screening and mainte-nance prioritization.
Zitieren
APA
(2026). Feature Importance-Based Machine Learning Classification of Distribution Trans-former Health Conditions Using Oil Test Data. International Journal of Advanced Research in Science and Technology, 15(7), 2169-2173. https://doi.org/10.62226/ijarst20262730
Vancouver
Feature Importance-Based Machine Learning Classification of Distribution Trans-former Health Conditions Using Oil Test Data. International Journal of Advanced Research in Science and Technology. 2026;15(7):2169-2173. doi:10.62226/ijarst20262730
BibTeX
@article{euvimed2026,
title = {Feature Importance-Based Machine Learning Classification of Distribution Trans-former Health Conditions Using Oil Test Data},
journal = {International Journal of Advanced Research in Science and Technology},
year = {2026},
volume = {15},
number = {7},
pages = {2169--2173},
doi = {10.62226/ijarst20262730}
}
EUVIMED