Random Forest Performance Analysis in Predicting Flashover Voltage of Polluted Insulators: A Comparative Study with PSO Application

Abstract

Insulator pollution constitutes one of the most significant challenges to the reliability of electrical power transmission networks, as it leads to reduced flashover voltage and increased failure probabilities. This study aims to evaluate the performance of the Random Forest algorithm in predicting the flashover voltage of polluted insulators by comparing two hyperparameter optimization methodologies: Grid Search and Particle Swarm Optimization (PSO). The analysis was conducted using a unified dataset consisting of 28 laboratory measurements and 140 computational samples, ensuring an objective comparison with previous studies. 

  The results demonstrated that the RF model optimized with Grid Search achieved balanced performance with high accuracy (RMSE = 0.1543) without signs of overfitting. In contrast, RF-PSO showed gradual improvement during training but failed to outperform RF-Grid on the test data, confirming the effectiveness of simple and systematic strategies for models with limited structure. These findings indicate that the choice of optimization technique should take into account the nature of the model and the characteristics of the dataset. This work provides practical evidence to guide researchers and engineers in adopting appropriate optimization strategies within electrical engineering applications.