Fast SVM incremental learning based on clustering algorithm

In the incremental learning process of support vector machines, the non-support vectors which is close to support vector samples are discarded in tradition method. But it is likely to change into the support vector after adding new training samples. To resolve this problem, this paper proposes a new method that combines support vector machine with clustering algorithm. In this method, firstly, use clustering algorithm to cluster the training sample set and get clustering particles ; secondly, look all centers of clustering particles as new samples training set and reconstruct the training samples set; then, train the new training samples set with fuzzy support vector machine (FSVM) and obtain the support vectors, and discard the samples that satisfy KKT conditions, put the samples that don not meet the KKT conditions and the support vectors together to reconstitute a new training set, train them again . Experimental results show that this method can enhance the classification accuracy rate and improve the speed of SVM training and classification speed, as keeping the generalization ability of SVM incremental learning.