A Hybrid Feature Selection Method for Software Defect Prediction

Abstract

Software Defect Prediction (SDP) is one of the important ways of software quality assurance, which uses the metric data to predict whether software module is defect. The quality of data influences the perfection of the prediction model. The high latitude containing some unnecessary features is one of the quality problem that dataset. To solve this problem, we proposed a hybrid feature selection (HFS) method combined different feature sorting technology. Firstly, we calculate the values of each feature include chi-squared (cs), Information gain (IG) and Pearson Correlation coefficient, respectively. Secondly, we sort the features based on the ranking of the three values to select features. Finally, we use the random forest to build the model. In order to validity the approach, we did experiments on 5 datasets in NASA. The result shows that our approach can select a smaller subset of features to improve the preformation in F-measure.