Effect of Data Sampling on Cone Shaped Embedded Normalization in Just in Time Software Defect Prediction

Just-in-Time (JIT) defect prediction represents a software engineering approach that seeks to detect potential defects in software code at the earliest stages of the development process. This proactive method allows developers to tackle issues before they escalate, thereby enhancing software security and reliability. However, researchers often encounter a common challenge known as class imbalance when working on this model. This imbalance in data adversely affects the model's performance. To address this, the study minimized the class imbalance problem by employing data sampling techniques. The study evaluated the performance of the proposed cone-shaped embedded normalization (CSEN) model against other baseline models in two scenarios. First, the comparison was conducted without sampling, and second, after performing data sampling. Typically, in state-of-the-art predictions of buggy changes, the f1 score ranges from 0.3 to 0.53. However, the proposed model significantly improved this score to 0.72. Moreover, the highest accuracy achieved by the proposed CSEN model was 74.42%.