An Ensemble Learning Algorithm Based on Resampling and Hybrid Feature Selection, with an Application to Software Defect Prediction

In recent years, software defect prediction(SDP) based on ensemble learning has attracted extensive attention. Hevertheless, the forthcoming integrated learning means still have some shortcomings in improving the divergence of base classifiers. How to improve the divergence of base classifiers is the major challenge in ensemble learning. In this paper, we advance the new ensemble learning algorithm named E_RHFS, and apply it to SDP. In E_RHFS, we first need the weighted complexity-based SMOTE to manage imbalanced data; Second, the resampling technology is used to generate multiple sampling sets from the balanced data. Moreover, for each sampling set, a hybrid feature selection method is used to select features, which combines the granular decision entropy and the Random Subspace Method (RSM); Third, a base classifier is constructed on each sampling set; Finally, all base classifiers are combined by voting, so as to obtain the final ensemble classifier for software defect prediction. Test on many relevant data sets express that E_RHFS be able to get better performance than existing ensemble methods.