A Systematic Literature Review on Machine Learning for Automated Requirements Classification

The development of quality software begins with the correct identification of the system needs. These requirements represent the basis of the subsequent activities in the software life cycle. The correct identification of these requirements in their different categories impacts on the actions taken to meet them. However, this classification can be often time-consuming or error-prone when it comes to large-scale systems, so different proposals have been made to assist in this process automatically. This systematic literature review identifies those applications of Machine Learning techniques in the classification of software requirements. In this regard, 13 articles were identified, from which relevant information on the applied algorithms, their training process, and their evaluation metrics are analyzed. From the results obtained, it is identified that the most recurrent classification algorithms featured on the identified studies are Naive Bayes, Decision Trees, and Natural Language Processing algorithms. The most frequent training datasets are academic databases and collected user reviews.