Pattern Recognition: A Comparison Between Support Vector Machines and Relevance Vector Machines

Abstract

This work deals with the comparison between support vector machines (SVM) and relevance vector machines (RVM) classifiers. Two data sets were generated from the numpy library, one predefined and the other randomly generated. The classifiers are trained using the datasets, with SVM using support vector classification (SVC) and RVM using expectation-maximization relevance vector classifier (EMRVC), both tested with the following kernels: RBF, polynomial, sigmoid, and linear. The main contribution of this work lies in observing the advantages and disadvantages of using each classifier. Given that the RVM model is Bayesian, it can provide a posteriori properties, but each classifier has its specific use.