Exploring the relationship between smells, bugs and harmful code through transfer learning

Abstract

The presence of code smells in a software project is a strong indication of its low quality in the context of its implementation, and in many cases, these same code smells can be harmful code segments for the application, becoming the culprits in bug generation. Currently, there are different approaches for detecting code smells, and there has been a recent deeper analysis of the correlation between these code smells and their harmfulness to the code. However, there is still much to be researched in the context of how we can improve the accuracy of detecting these potentially harmful codes. With this in mind, this work aims to expand the methods for detecting harmful code by using transfer learning to build a large dataset for training and validating machine learning models.