Abstract
The performance of software defect prediction heavily suffers from data-imbalance. In this article, the imbalance problem using SMOTE and resampling methods has been solved. The performance of software defects prediction with imbalance data and without imbalance data has also been studied. Experiments with WEKA 3.8.3 have been conducted and the performance of different classifiers are calculated using oversampling and resampling methods. Further, for statistical analysis paired T-TEST is used to validate the results. The effectiveness of oversampling and resampling methods for different classifiers are also checked.
The results show that after oversampling and resampling, the performance of classifiers has significantly increased and the winner classifiers are bagging AdaBoost and Random Forest in many cases.