Regresar

Requirements and GitHub Issues: An Automated Approach for Quality Requirements Classification

Abstract:

Abstract: In the development of quality software, critical decisions related to planning, estimating, and managing resources are bound to the correct and timely identification of the system needs. In particular, the process of classifying this customer input into software requirements categories tends to become tedious and error-prone when it comes to large-scale systems. On the ground described by a complementary systematic literature review, this research introduces a proposal on the application of Machine Learning techniques for automated software requirements classification. In this regard, the training and later hyperparameter optimization through Differential Evolution of five classification models are carried out based on quality attributes examples found in the available literature. As a case study, these models are tested with issue reports collected from five open-source projects at GitHub to identify quality-attributes-related knowledge on such user feedback. The finding of the most characteristic terms by quality attribute through the TF-IDF algorithm stands out from the training. The results show a moderately high ability to classify other generic software requirements correctly, achieving a Geometric Mean of up to 82.51%. However, the same classifiers applied to issue reports showed significant difficulties identifying information related to quality attributes, since an F-Score no greater than 50% was reached.