Machine Learning Driven Guidance for Software Maintenance: Enhancing Code Management and Features

Abstract

Software systems undergo continuous maintenance to fix defects, enhance features, and improve code quality to ensure that systems stay reliable and relevant in a competitive market. Due to the escalating growth of the code base and the increasing complexity of software systems, it is essential to adopt automated approaches that support and drive software maintenance process. Despite existing research efforts using machine learning (ML) techniques to improve software maintenance, developers continue to encounter challenges in managing code changes and clone evolution in code maintenance and enhancing features. To assist developers in maintaining software effectively, in this thesis, we introduce a set of approaches that leverage software artifacts and ML to assist developers with two key areas of software maintenance: managing code and enhancing features. Specifically, we conduct four studies: (1) improving the accuracy of predicting change impact for defect resolution by utilizing common characteristics of issue reports; (2) analyzing the dynamics and impact of code clones in deep learning frameworks to ensure long-term code quality and maintainability; (3) performing automatic, comprehensive competitor feature analysis, and (4) generating suggestions for feature improvements based on competitor user review analysis. The aim of this thesis is to empower developers with novel methodologies for streamlined and effective software maintenance, thereby fostering a high-quality maintained codebase and competitive, sustainable software systems. The approaches proposed in the thesis can be adapted to software projects where mobile user reviews, source code, and issue reports are available to demonstrate the applicability of the research contributions.