Comparative analysis of optimization models for the binary classification problem by the SVM method

Abstract

In the article several optimization models are considered for solving the binary classification problem by the support vector machine method. Mathematical statements of models differ in the form of the objective function and variables responsible for the magnitude of the classification error for each example of the training set. The main goal is to change the general principle of the formulation of the objective function, making the main criterion for minimizing the error function. Multicriteria statements of the optimization model are also considered. The article presents a conceptual and computational comparison of the suggested models of optimization problems with a computational experiment that estimates the time of the solution of the problem, the number of iterations, the received accuracy of classification. The comparison is performed on known data sets formed for the support vector machine method.