Mining Software Bug Repositories: A 15-Year Analysis of Trends, Techniques, and Limitations in Bug Localization, Classification, Triaging, and Resolution (2010–2024)

Abstract

Over the past decade and a half, mining software bug repositories has become a critical research domain for improving software maintenance, quality assurance, and automated debugging processes. This study presents a comprehensive analysis of bug localization, classification, triaging, and resolution trends from 2010 to 2024, based on data from prominent repositories such as Eclipse, Mozilla, KDE, Apache, and OpenStack. Using a dataset comprising over 100 peer-reviewed research papers and repository-derived performance metrics—including Mean Average Precision (MAP), F1-score, and Mean Reciprocal Rank (MRR)—the research identifies key methodological advances and persistent challenges in automated bug handling. The findings reveal that machine learning–driven approaches, particularly deep learning models, have significantly improved classification accuracy, often exceeding 90%, while hybrid techniques integrating textual, contextual, and developer history data have reduced bug triaging delays. However, bug localization remains hindered by imbalanced and noisy data, and resolution automation suffers from limited dataset standardization and cross-repository generalizability. Temporal trends indicate a shift from rule-based methods to multi-modal AI frameworks, leveraging natural language processing, statistical modeling, and repository mining. This work contributes a synthesized understanding of the field’s evolution, highlights gaps such as inconsistent reporting formats and lack of explainable AI adoption, and provides recommendations for future research aimed at developing standardized, scalable, and interpretable bug management solutions.