Sample selection based on maximum entropy for support vector machines

Abstract

It is always true that in the classification problems, unlabeled data is abundant while the cost for labeling data is expensive. In addition, large data sets often contain redundancy hence degrade the performance of the classifiers. In order to guarantee the generalization capability of the classifiers, a certain number of suitable unlabeled samples need to be selected out and labeled. This process is referred to as sample selection. In this paper, we propose an active learning model of sample selection for support vector machines based on the measurement of neighborhood entropy. In order to evaluate the capability of the generated SVMs, experiments have been conducted on several benchmark data sets. Comparisons between our proposed method and the random selecting method have also been conducted.