Abstract

The quality of software depends heavily on how accurately it works. The accuracy is determined by the fact that the less the software modules are defect prone, the more accurate the software will be. So software defect prediction which classifies software modules into defect prone and non-defect prone categories is an important area where a lot of research works are being done. Cost sensitive learning that has been adopted in software defect prediction aims to minimize total expected cost. In this paper a two-stage cost sensitive learning is proposed where the cost information is used in the feature selection stage and in the classification stage. Three cost sensitive algorithms, Cost-Sensitive Variance Score, Cost-Sensitive Laplacian Score, and Cost-Sensitive Constraint Score are proposed. The results of the proposed methods are analyzed with datasets from NASA.

Authors and Affiliations