Classification Based on Clustered Group SVM

A novel algorithm which combines clustering analysis and SVM is proposed for classification. Specifically, based on the conglomeration and decentralization characteristics of the positive and negative samples, we present a new type of support vector machine called Clustered Grouping Support Vector Machine or GC-SVM. After clustering training, the samples are divided into different groups, then a series of SVM sub-classifiers are designed for each other two groups. During testing stage, a sample is classified based on one of the sub-SVM classifiers according to the weighted Euclidean distance from the center of the clusters. Our algorithm incorporates the merits of both SVM and cluster analysis, which casts a difficult two-class problem into a set of simple optimized multi-classifiers. The experimental results show that the performance of the proposed algorithm outperforms the traditional SVM method.