Cone vertex estimation in polyhedral conic classifiers

Recently, polyhedral conic classifiers have become popular since they perform better compared to the Support Vector Machines (SVMs). Cone vertex of polyhedral conic classifiers is an important parameter and it is generally taken as the mean of positive data in literature. In this paper, we studied optimally estimating the cone vertex to improve the accuracy of the polyhedral conic classifiers. The proposed method has been compared with the linear SVMs and polyhedral conic classifiers and extended polyhedral conic classifiers that fix the cone vertex to the mean of the positive data. Various real databases in UCI Machine Learning Repository and a synthetic data set have been used for comparison of these classification methods. Experimental results show that the proposed approach has good results in solving data classification problems.