A Hybrid Text Classification Method Based on K-Congener-Nearest-Neighbors and Hypersphere Support Vector Machine

Abstract

Our work implements a novel text classifier by combining k congener nearest neighbors-Support Vector Machine(KCNN-SVM) with hyper sphere Support Vector Machine(hyper sphere-SVM) training algorithm. Hyper plane Support Vector Machine has been widely used to divide the samples into two equal categories. However, the hyper sphere Support Vector Machine can not only separate the samples, but also divide them into two different parts. Since the probability inside and outside the hyper sphere is not same, hyper sphere-SVM is helpful to the classification when the datasets are imbalanced that we can control the radius of hyper sphere to get higher accuracy. The KCNN-SVM algorithm distinguishes a sample with its nearest neighbor's category label as well as the average distance between it and its k nearest same kind of neighbors which can enhance the accuracy when the samples are chaotic imbalanced. In this paper, we propose the hyper sphere-KCNN-SVM(HS-KCNN-SVM) hybrid approach which can validly improve the classification accuracy especially for those chaotic imbalanced samples.