An effective support vector machines (SVMs) performance using hierarchical clustering

Abstract

The training time for SVMs to compute the maximal marginal hyper-plane is at least O(N/sup 2/) with the data set size N, which makes it nonfavorable for large data sets. This work presents a study for enhancing the training time of SVMs, specifically when dealing with large data sets, using hierarchical clustering analysis. We use the dynamically growing self-organizing tree (DGSOT) algorithm for clustering because it has proved to overcome the drawbacks of traditional hierarchical clustering algorithms. Clustering analysis helps find the boundary points, which are the most qualified data points to train SVMs, between two classes. We present a new approach of combination of SVMs and DGSOT, which starts with an initial training set and expands it gradually using the clustering structure produced by the DGSOT algorithm. We compare our approach with the Rocchio Bundling technique in terms of accuracy loss and training time gain using two benchmark real data sets.