This paper develops a hybrid binary particle swarm optimization (BPSO) and evolutionary algorithm (EA) based feature selection method. Inspired by the concept of binary PSO, the particle's position updating process is designed in a binary search space. The fitness function is defined as the accuracy of the ENN classifier. The feature selection method using a hybrid BPSO-EA learning algorithm is developed and described. The experiments include the comparison of ENN classification accuracy with and without the BPSO-EA feature selection method. The feature reduction rate between the proposed BPSO-EA-ENN method and the BPSO+C4.5 method is also compared. In addition, a comparison of BPSO-EA-ENN to other classification methods is provided. The experimental results demonstrate that the proposed BPSO-EA feature selection method improves the classification accuracy. In addition, our proposed method has higher improved accuracy and feature reduction rate than the BPSO+C4.5 feature selection method on the Ionosphere data set, as well as better accuracy rate than the BPSO+C4.5 method on the Movement Libra data set. Further, the overall classification accuracy of our proposed BPSO-EA-ENN outperforms ENN, KNN, Na?ve Bayes, and LDA classification methods on the eight UCI data sets.
展开▼
