A Reduction SVM Classification Algorithm Based on Adaptive AP Clustering Granulation

Abstract

The classification speed of SVM is inversely proportional to the number of Support Vectors (SVs). Therefore, the less SVs means the more sparseness and the higher classification speed. In order to reduce the number of SVs but without losing of generalization performance, a new algorithm called Classification Algorithm of Support Vector Machine based on Adaptive Affinity Propagation clustering Granulation (CSVM-AAPG) is proposed, which employs Affinity Propagation (AP) clustering algorithm to cluster the original SVs and the cluster centers are used as the new SVs, then aiming to minimize the classification gap between SVM and CSVM-AAPG, a quadratic programming model is built for obtaining the optimal coefficients of the new SVs. Meanwhile, it is proven that when clustering the original SVs, the minimal upper bound of the error between the original decision function and the fast decision function can be achieved by AP. Finally, experiments show that compared with original SVs, the number of SVs decreases and the speed of classification increases using CSVM-AAPG, while the loss of accuracy is in the acceptable level.