<![CDATA[This study analyzes the performance of electricity consumption prediction by comparing a metaheuristic optimization-based Artificial Neural Network (ANN) model from previous studies with a standalone Support Vector Regression (SVR) model. The SVR model uses an ε-insensitive linear regression approach with regularization, and feature normalization is performed to maintain training stability. Experimental results show that SVR is capable of producing very high prediction accuracy on training data with a relative accuracy per sample of 96.74%–99.90% and an average training accuracy of 99.09%, while maintaining generalization on test data with an accuracy of 92.56%. Compared to ANN with metaheuristic optimization (GA/PSO), which generally requires a more complex training process and relies on initialization, SVR offers the added value of more stable, deterministic, and easily reproducible training with lower model complexity. These findings confirm that SVR can be an effective and efficient alternative for electricity consumption prediction as well as a strong comparison to the ANN metaheuristic approach.]]>
Copyrights © 2026
