Software defect prediction using fuzzy support vector regression

Abstract

Regression techniques have been applied to improve software quality by using software metrics to predict defect numbers in software modules. This can help developers allocate limited developing resources to modules containing more defects. In this paper, we propose a novel method of using Fuzzy Support Vector Regression (FSVR) in predicting software defect numbers. Fuzzification input of regressor can handle unbalanced software metrics dataset. Compared with the approach of support vector regression, the experiment results with the MIS and RSDIMU datasets indicate that FSVR can get lower mean squared error and higher accuracy of total number of defects for modules containing large number of defects. © 2010 Springer-Verlag.