SCALABLE ENSEMBLE-BASED BUG PREDICTION MODEL FOR MODERN SOFTWARE ENGINEERING ECOSYSTEMS

Abstract

Software maintenance in large open-source environments relies heavily on efficient and accurate bug triage. Projects such as Eclipse generate tens of thousands of issue reports each year across diverse components and severity levels, making manual classification slow, inconsistent, and difficult to scale. Earlier automation efforts using single machine learning models like SVM and Logistic Regression have achieved only moderate accuracy and often struggle to adapt to evolving bug datasets. This work proposes an ensemble-driven, scalable framework for automated Eclipse bug classification. The system processes raw bug descriptions through preprocessing steps including tokenization, stop-word removal, and lemmatization, followed by TF-IDF-based feature extraction. Five models—SVM, Random Forest, Logistic Regression, Extra Trees Voting ensemble, and XGBoost—are trained and evaluated on a curated Eclipse–Mozilla dataset. A user-friendly GUI also enables non-expert users to upload data, visualize preprocessing, and select models. Under a 70/30 train–test split, results show improved performance, with XGBoost achieving the highest scores at 92.27% accuracy, 92.91% precision, 92.65% recall, and 92.50% F1-score, demonstrating the effectiveness of the proposed ensemble-oriented approach.