Application of Deep Learning for Code Smell Detection: Challenges and Opportunities

Code smells are indicators of deeper problems in source code that affect the system maintainability and evolution. Detecting code smells is crucial as a software maintenance task. Recently, there has been a growing interest in utilizing deep learning techniques for code smell detection. However, there is limited research on the current state-of-the-art in this topic. To bridge this gap, this paper conducts a systematic literature review to investigate the application of deep learning in code smell detection. We have followed a well-defined methodology for conducting a systematic literature review in the field of software engineering. Through this process, we have identified a total of 30 primary studies. The reviewed studies have been thoroughly analysed according to different research aspects in the detection process, including the used code representations and learning algorithms. We have also explored the frequently identified code smells in terms of type and number, where the results have revealed a research gap in this context. Our analysis has also focused on determining how detection performance was evaluated and validated. Specifically, we have compiled a list of available datasets and compared between them based on various criteria. It was observed that the oracles within these datasets were developed using different approaches, but supported a limited number of identified code smells. Furthermore, through our investigation, we have highlighted several challenges that need to be addressed. Alongside the challenges, we have identified numerous promising opportunities that serve to improve future research.