Classifier Evaluation for Software Defect Prediction

Gang Kou, Yi Peng, Yong Shi, Wenshuai Wu

Abstract

Feature selection is an essential step in the process of software defect prediction due to the negative effect of irrelevant features on classification algorithms. Hence selecting the most relevant and representative features is critical to the success of software defect detection. Another problem in software defect prediction is the availability of a large number of classification models. This paper applies feature selection and classifier evaluation in the context of software defect prediction. An empirical study is presented to validate the proposed scheme using 9 classifiers over 4 public domain software defect data sets. The results indicate that the proposed scheme can improve the performance of classifiers using the most representative features and recommend classifiers that are accurate and reliable in software defect prediction.

Keywords

software defect detection; classifier evaluation; multi-criteria decision making (MCDM);

View full article