Leveraging CatBoost and XAI for Enhanced Software Defect Prediction

Software defect prediction (SDP) plays a crucial role in identifying faulty modules within software systems, enhancing their reliability, and reducing maintenance costs. Till now, a number of machine learning (ML), deep learning (DL), and ensemble methods have been applied for software defect prediction. However, there are other ensemble methods that are yet to be explored in SDP. Moreover, the defect datasets are highly imbalanced which reduces the accuracy of the models. To address these issues, we utilized a CatBoost classifier to predict software defects and the Synthetic Minority Over-sampling Technique (SMOTE) to address the class imbalance problem. In addition, for better interpretability of the defect features of the datasets, we employed three explainable AI techniques: Local Interpretable Model-agnostic Explanations (LIME), Shapley Additive Explanations (SHAP), and Partial Dependency Plot (PDP). To evaluate the performance of the proposed approach, we compared our results with the existing ML models where CatBoost achieved better F1 scores, accuracy, precision, and recall values than existing ML models. Furthermore, all three explainable AI techniques demonstrated that feature ev(g) or McCabe’s "essential complexity" contributes most towards the classification of the model.