A Hybrid Prototype Construction and Feature Selection Method with Parameter Optimization for Support Vector Machine

Abstract

In this paper, an order-independent algorithm for data reduction, called the Dynamic Condensed Nearest Neighbor (DCNN) rule, is proposed to adaptively construct prototypes in training dataset and to reduce the over-fitting affect with superfluous instances for the Support Vector Machine (SVM). Furthermore, a hybrid model based on the genetic algorithm is proposed to optimize the prototype construction, feature selection, and the SVM kernel parameters setting simultaneously. Several UCI benchmark datasets are considered to compare the proposed GA-DCNN-SVM approach with the GA-based previously published method. The experimental results show that the proposed hybrid model outperforms the existing method and improves the classification accuracy for SVM.