Study of integration method based on dynamically selected Support Vector Machine and simulation

This paper proposes a Support Vector Machine integration methodology, which is based on dynamic selection method. Using FCM-SD algorithm, an improved method from FCM and closeness degree algorithms, the effective neighborhood of the discriminated sample is determined. Furthermore, based on the accuracies of segment classifications, a set of optimal individual classifiers are selected. Finally a weighted majority vote method is used to integrate the selected classifiers. Simulation results show that the proposed method reduces the complexity of the Integrated Classification model. It effectively improves the classification performance as well.