Comparative Analysis for Probability Modeling of Multi-class SVM

摘要

The one-against-one method and the one-against-rest method are two popular multi-class classification methods that combine together all results of two-class support vector machine classifiers. The paper presents the probability output of two methods in multi-class SVMs. The binary output and the probability output of two multi-class SVM methods in terms of classification precision and the total times of both training and predicting stages are compared and analyzed in order to evaluate the classification performance of probability output of multi-class SVM. Three experiment results show that the probability output of the one-against-one method exhibits the excellent classification performance in terms of classification precision and computational cost