Hybrid optimization of feature selection and SVM training model

A method was developed for hybrid optimization of feature selection and support vector machine (SVM) training models, using the interdependent relationship between the feature selection and the training model to improve the SVM performance. The method includes three important techniques: (1) The objective function is the SVM performance which is calculated using the ξα-estimate method; (2) The feature selection is described by a binary vector and the training model is described by a mixed kernel and tradeoff control parameters; (3) The hybrid optimization problem is solved using evolutionary algorithms. A standard data set for intrusion detection was used to compare the hybrid optimization method with single optimization and separate optimization methods and to compare the performances using genetic algorithms (GA) and particle swarm optimization (PSO). The experimental results show that the hybrid optimization method guarantees better SVM performance and that the optimization process using this approach has a higher rate of convergence than the other optimization methods.