Leveraging Large Language Models (LLMs) for Automated Bug Detection and Resolution in Software Engineering

Automated detection and resolution of software bugs are essential in software engineering due to the growing complexity of software systems. Debugging tasks traditionally require substantial resources. This research explores using Large Language Models (LLMs), specifically GPT-based models, to automate bug detection, classification, and resolution. A novel workflow introduced using GPT-4, systematically analyzing software bug reports and source code snippets. Bugs are classified into categories like syntax errors, logic errors, and performance issues. The model then proposes targeted fixes based on historical resolution data. Experimental results using the Defects4J dataset showed significant improvements compared to manual debugging, achieving a detection accuracy of 89.7% and resolution efficacy of 86.4%. These outcomes underline the potential of LLMs to reduce manual debugging effort, improve software maintenance efficiency, and enhance software quality.