Implementation of Random Forest and Support Vector Machine Models for Predictive Failure Detection in Power Transmission Equipment

Abstract

Unplanned failures of power transmission equipment cause costly outages and safety hazards, motivating a shift from reactive and time-based maintenance to predictive maintenance (PdM). This study implements and compares two supervised machine learning models, Random Forest (RF) and a Radial-Basis-Function Support Vector Machine (SVM), to predict the likelihood of failure over a 30-day horizon across five classes of transmission equipment: transformers, circuit breakers, insulators, towers, and conductors. We assembled a condition-monitoring dataset of 6,000 records from four representative Transmission Company of Nigeria (TCN) substations, capturing electrical, thermal, mechanical, dissolved-gas, and environmental parameters. We derived failure labels from a composite risk score grounded in established engineering thresholds (IEEE C57.104/IEC 60599 dissolved-gas limits, partial-discharge severity, aging curves, and thermal loading), yielding a realistic 23.22% failure prevalence (3.31:1 class imbalance). Realistic sensor dropout (2.91% missingness), calibration glitches, and duplicate telemetry were simulated and resolved through a seven-step pre-processing pipeline combining group-wise imputation, missingness-indicator flags, and conservative winsorization. Historical failure count, equipment age, corrosion index, and dissolved-gas concentrations emerged as the dominant predictors, consistent with established transmission asset degradation theory. Both models were implemented in scikit-learn, validated with 5-fold stratified cross-validation, and evaluated on a held-out test set of 1,200 records. Random Forest outperformed SVM on every metric: accuracy (81.3% vs 75.2%), precision (62.7% vs 46.8%), F1-score (54.8% vs 48.4%), ROC-AUC (79.8% vs 72.3%), and average precision (59.1% vs 46.7%). The study concludes that Random Forest is the more suitable of the two models for transmission equipment failure prediction and recommends its deployment as a risk-ranking and maintenance prioritization tool.