A Provenance-Centric Approach to Code Smell Reduction using Human-in-Loop Approach

Code smells are early warning signs of software development whose detection and timely resolution are essential for long-term software quality. Conventional methods for detecting smells fail to complete the task because they do not consider the evolution of code over time. This paper suggests a provenance-driven approach to integrate a human-in-the-loop approach in code smell detection to overcome this issue. The study begins by collecting provenance data from software version control systems. This data includes a history of code changes, decisions and developer feedback. Secondly, automated tools like PMD and FindBugs detect code smells. The results generated by these tools are reviewed and validated by developers to refine the findings. The process works in the form of a feedback loop, which brings continuous improvement to the method. The system analyzes the provenance data to understand the context of smells and identify the reason behind smell emergence. Finally, we generate refactoring suggestions based on this analysis. The proposed method is compared with SonarQube, Checkstyle, PMD and FindBugs to evaluate the performance using various metrics. Results show that the proposed approach reduces false positives by 30% and increases the detection of complex smells by 25%, boosting software maintainability.