Comparative study of Sampling Techniques for Software Defect Prediction

SDP has been an evolved and important step in the Software Development Life Cycle during the testing phase. Significant research has gone through the prediction of defects thereby enhancing the quality of software produced quickly and efficiently. SDP datasets are quite imbalanced, and this has caused the predictive models to suffer in terms of performance. In this research we have observed and studied the impact that Over Sampling Technique like Synthetic Minority Over Sampling Technique and Under Sampling Technique like Random Under Sampling Technique have on the prediction of software defects. Useful features were extracted from the datasets using Random Forest, then, we have implemented ensemble classification models such as AdaBoost and Bagging Classifier, in addition to that we have also observed the predictive nature of Support Vector Machine. Finally, we compared the results obtained between the various models, the different sampling techniques, using performance measures like El-Score, Accuracy, Area under the Receiver Operating Curve as well as G-Mean.