An empirical study toward dealing with noise and class imbalance issues in software defect prediction

The quality of the defect datasets is a critical issue in domain software prediction (SDP). These are obtained through mining repositories. Recent studies claim over dataset. It because inconsistency between bug/clean fix keyword fault reports and corresponding link change management logs. Class Imbalance (CI) problem also big challenging SDP models. method trained using noisy imbalanced data leads to inconsistent unsatisfactory results. Combined analysis instances CI needs be required. To best our knowledge, there insufficient that have been done such aspects. In this paper, we deal with impact noise on five baseline models; manually added various level (0–80%) identified its performance those Moreover, further provide guidelines for possible range tolerable We suggested model, which has highest ability outperforms other classical methods. True Positive Rate (TPR) False (FPR) values models reduce 20–30% after adding 10–40% instances. Similarly, ROC (Receiver Operating Characteristics) 40–50%. model avoid 40–60% as compared traditional