Experimentation of Code Smells Using Deep Learning Techniques
Omeir Fawaz, Mohammad Amaan, Suryansh Sahu, Mohd Adnan, Aakanshi Gupta · 2023
A code smell is a measurable indicator that highlights significant issues in the software development process caused by inadequate programming practices. These are usually introduced in a software program during the design or implementation phases, and may lead to more significant problems during software maintenance. The existing methods to detect smells based on metrics, heuristics or machine learning have certain limitations. In this paper, we suggest two models built on deep learning for predicting code smells: convolutional neural network (CNN) and recurrent neural network (RNN). These models use features extracted from source code to detect six different code smells. We then evaluate and compare their results. The proposed approach performs remarkably well for single code smell identification according to the experimental results obtained.