Generalization enhancement of support vector regression in electric load forecasting with model selection

Abstract

Listen

Translate article

Nowadays, feature selection and parameter optimization are two fundamental issues in machine learning. The former improves the quality of the algorithm by selecting a subset of features, while the latter concerns finding the most suitable parameter values. The two issues have the same objective of improving the predictive performance of the algorithm. In particular, machine learning algorithms capacity can be straightened using particle swarm optimization to avoid the problem of overfitting [3] by both feature selection and parameter optimization. This paper focuses on the restriction of this general issue to the support vector regression (SVR) algorithm and the electric load forecasting problem. That is, feature selection and the optimal parameter setting have been considered simultaneously in order to further enhance it generalization capacity. Experimental results on a widely used electric load dataset show that our proposed hybrid method for model selection by both feature selection and parameter optimization of SVR can achieve better generalization capacity when compared with the classical SVR model while using feature selection and without using it.