An Enhanced Software Bug Prediction Framework Using Machine Learning Model

: Software bug prediction is an essential process that significantly improves software quality, reduces development costs, and optimizes resource allocation. Organizations can prevent costly errors and improve system reliability by identifying defective software modules early in the development life cycle. In recent years, combining machine learning techniques and ensemble learning approaches has enhanced the accuracy and performance of software bug prediction models. This study proposes a robust stacked ensemble-based approach for software bug prediction, combining the predictive capabilities of Random Forest, XGBoost, and CatBoost as base classifiers with a neural network meta-model to improve the decision-making process. To ensure robust model performance, hyper-parameter optimization will be applied, and the dataset will be carefully pre-processed, including class balancing with ADASYN, feature selection using Recursive Feature Elimination (RFE), and data augmentation via noise injection. The approach will be evaluated on different prominent datasets of NASA and Promise dataset. The proposed model will be evaluated based on accuracy, precision, recall, and F1-scores. The experimental results will be compared with the current top-performing models. Utilizing the strengths of ensemble learning and systematic optimization, our proposed methodology may demonstrate a reliable solution for software defect prediction.