A Survey on Data Science Techniques for Predicting Software Defects

Abstract

In recent years, data science has been used extensively to solve several problems and its application has been extended to several domains. This paper summarises the literature on the synergistic use of Software Engineering and Data Science techniques (e.g. descriptive statistics, inferential statistics, machine learning, and deep learning models) for predicting defects in software. It shows that there is a variation in the use of data science techniques and limited reasoning behind the choice of certain machine learning models but also, in the evaluation of the obtained results. The contribution of this paper has to be intended as a categorization of the literature according to the most used data science concepts and techniques from the perspectives of descriptive and inferential statistics, machine learning, and deep learning. Furthermore, challenges in software defect prediction and comments on future research are discussed and forwarded.