An improved incremental training algorithm for support vector machines using active query

摘要

In this paper, we present an improved incremental training algorithm for support vector machines (SVMs). Instead of selecting training samples randomly, we divide them into groups and apply the k-means clustering algorithm to collect the initial set of training samples. In active query, we assign a weight to each sample according to its confidence factor and its distance to the separating hyperplane. The confidence factor is calculated from the error upper bound of the SVM to indicate the closeness of the current hyperplane to the optimal hyperplane. A criterion is developed to eliminate non-informative training samples incrementally. Experimental results show our algorithm works successfully on artificial and real data, and is superior to the existing methods.