Software Defect Prediction Using Dynamic Support Vector Machine

Abstract

In order to solve the problems of traditional SVM classifier for software defect prediction, this paper proposes a novel dynamic SVM method based on improved cost-sensitive SVM (CSSVM) which is optimized by the Genetic Algorithm (GA). Through selecting the geometric classification accuracy as the fitness function, the GA method could improve the performance of CSSVM by enhancing the accuracy of defective modules and reducing the total cost in the whole decision. Experimental results show that the GA-CSSVM method could achieve higher AUC value which denotes better prediction accuracy both for minority and majority samples in the imbalanced software defect data set.