C-PNN: Multiple-instance Learning Based on Clustering and Probabilistic Neural Network

Abstract

Due to a significant difference between single-instance learning(SIL) and multiple-instance learning(MIL), many traditional learning methods such as nearest neighbor classification, SVM and other classification algorithms are difficult to directly apply on MIL problem. Although many MIL algorithms have been proposed in recent decades, the complexity of most is high. To solve this problem, a new algorithm based on clustering and probabilistic neural network is proposed in this article. Experimental results show that the proposed approach not only can obtain a nearly good classification accuracy in classical UCI data sets, but also the algorithm complexity is reduced in comparison with other methods at the same time.