Bias And Variance Multi-Objective Optimization For Support Vector Machines Model Selection

In this paper, we describe a novel model selection approach for a SVM. Each model can be composed by a feature selection method and a pre-processing method besides the classifier. Our approach is based on a multi-objective evolutionary algorithm and on the bias-variance definition. This strategy allows us to explore the hyperparameters space and to select the solutions with the best bias-variance trade-off. The proposed method is evaluated using a number of benchmark data sets for classification tasks. Experimental results show that it is possible to obtain models with an acceptable generalization performance using the proposed approach.