A New Formulation for Cost-Sensitive Two Group Support Vector Machine with Multiple Error Rate

Abstract

Support vector machine (SVM) is a popular classification technique which classifies data using a max-margin separator hyperplane. The normal vector and bias of the mentioned hyperplane is determined by solving a quadratic model implies that SVM training confronts by an optimization problem. Among of the extensions of SVM, cost-sensitive scheme refers to a model with multiple costs which considers different error rates for misclassification. The cost-sensitive scheme is useful when misclassifications cannot be considered equal. For example, it is true for medical diagnosis. In such cases, misclassifying a patient as healthy implies more loss in comparison to the opposite loss. Therefore, cost-sensitive scheme poses as a modified model and hereby aims at minimizing loss function instead of generalization error. This paper, concentrates on a new formulation cost-sensitive classification considering both misclassification cost and accuracy measures. Also, in the training phase a new heuristic algorithm will be used to solve the proposed model. The superiority of the novel method is affirmed after comparing to the traditional ones.