ReBaRF-Bug: A Multi-Stage Augmentation and Deep Learning-Enhanced Approach for Automated Bug Classification and Prioritization
Deepshikha Chhabra, Raman Chadha · 2025
Automation of bug classification and prioritization is essential for software maintenance. Unfortunately, traditional techniques fail to solve the high imbalance and feature sparsity problems intextual bug reports. This work suggests ReBaRF – Bug, a novel approach using advanced Natural Language Processing (NLP), feature augmentation, profound learning-based feature transformation and ensemble learning for automated bug categorization and priority prediction. Firstly, preprocessed bug reports collected from repositories of Eclipse and Mozilla (2016–2019) are fed into a TF-IDF-based feature extraction process for data preparation. To remedy the class imbalance, a multi−stage augmentation technique combining SMOTE, paraphrasing, and class−weighted oversampling is used to improve classification stability. For feature transformation of textual features before classification, the Residual Neural Network (ResNet) is used, and Random Forest with Bagging and Boosting is used for classification. By achieving 94.5% for Eclipse and 95.0% on Mozilla, the proposed ReBaRF-Bug model can outperform other models such as Random Forest (92.3% for Eclipse, 92.7% for Mozilla) and Decision Tree (88.94% for Eclipse, 89.32% for Mozilla). Additionally, its improvements in precision, recall, and F1 score are shown to validate its robustness, while ROC AUC scores of 95.3% (Eclipse) and 96.0% (Mozilla) prove that its classification boundaries are much better. Augmentation contributes to the additional performance improvement, increasing the accuracy from 88.9% to 94.5% (Eclipse) and 89.4% to 95.0% (Mozilla).ReBaRF Bug exploits ResNet for feature transformation, does multi level augmentation with a novel hybrid feature engineering, and thus generalizes and adapts to various bug reports. Adopting this approach reduces manual triaging efforts.The proposed method creates a benchmark for automation and prioritization, thus creating a scalable and reliable solution to automatic bug triaging.