Bug Severity and Priority Prediction using Machine Learning Techniques

Abstract— Manual bug triaging in large software projects is time-consuming, subjective, and difficult to scale as bug reports increase. [1] Accurate severity and priority assignment, fix-time estimation, and developer allocation are essential to avoid maintenance delays and missed releases. [2] This paper compares two machine learning approaches for automating bug triage using a real-world Mozilla Bugzilla dataset. [3]Both methods are evaluated on the same cleaned and balanced data for fair comparison. The first is a hybrid end-to-end system using TF–IDF features, LightGBM models, and semantic similarity to predict priority, estimate fix time, and recommend developers. The second is a simpler and more interpretable pipeline using TF–IDF with SVM and Logistic Regression to predict severity and priority. While the hybrid model achieves stronger overall performance, the lightweight approach offers competitive accuracy with lower complexity and easier deployment, highlighting key trade-offs between performance and practicality. Keywords— Bug Triage, Severity Prediction, Priority Prediction, Machine Learning, Software Maintenance