Abstract:

The generalization performance of SVM applied to classification problems will be reduced if different class data are seriously overlapped. A new approach EBSVM (entropy based support vector machine) is presented to prune data based on the concept of the information entropy for support vector machine. The EB-SVM employs the information entropies of the training data to remove the patterns far from the boundaries and delete the noise and overlapped instances close  to the boundaries, and then uses the pruned dataset to construct a SVM classifier. Experimental results show the EB-SVM takes less time than SVM and improves the classification accuracy.

Key words: information entropy; data pruning; support vector machine; classification; data distribution