Performance Analysis of Random Forest Algorithm For Health Index Classification Of PLN Distribution Transformer

Abstract

The rapid growth in Indonesia's electricity demand has increased the urgency for effective asset management in PLN’s distribution network, particularly for distribution transformers. As critical infrastructure, these transformers require accurate condition monitoring to reduce operational risks and optimize maintenance planning. Although PLN has adopted a Health Index (HI) system using parameters such as oil condition and load current, its maintenance strategy remains largely schedule-based and lacks strong data-driven support. This study addresses the gap by applying the Random Forest (RF) algorithm to classify transformer health conditions based on historical inspection data. Random Forest is an ensemble machine learning algorithm that builds multiple decision trees and merges their results to improve predictive accuracy and control overfitting. It is especially suitable for classification tasks involving complex and imbalanced datasets, making it an ideal choice for utility asset diagnostics. The objective is to develop an accurate, interpretable, and scalable model that supports the transition to condition-based maintenance. Performance evaluation was conducted using two data partitioning schemes: 70% training, 15% validation, and 15% testing; and 80% training, 10% validation, and 10% testing. The RF model consistently outperformed Logistic Regression across all metrics, achieving up to 99.54% accuracy, 99.53% precision, 99.54% recall, and 99.52% F1-score. It also demonstrated lower Mean Squared Error (MSE)—0.0194 and 0.0135—compared to Logistic Regression’s 0.0575 and 0.0613. Feature importance analysis identified oil condition and load current as key predictors, while spatial visualization enabled intuitive geo-mapping of transformer health status. These results demonstrate that Random Forest is robust in handling imbalanced utility data and effective for practical deployment in asset health classification. The findings offer valuable insights to support PLN’s digital transformation toward predictive maintenance and intelligent asset management.