Regression Test Generation by Usage Coverage Driven Clustering on User Traces
Frédéric Tamagnan, Fabrice Bouquet, Alexandre Vernotte, Bruno Legeard · 2023
Regression testing is used to verify that changes or modifications do not negatively impact existing features or functionality. However, it can be a significant bottleneck due to the challenge of determining the optimal number of tests for adequate coverage and fast execution and the difficulty for Quality Assurance engineers to anticipate end-user behavior. This paper presents a Usage Coverage Driven Clustering (UCDC) approach that regroups user trace logs in clusters of similar behavior, then constitutes the optimal test suite by electing one user trace per cluster as test cases. To find the optimal number of clusters, a novel metric is proposed to evaluate the statistical representativeness of the selected test candidates. Experimentation on three datasets1shows that the proposed metric, as compared to generic metrics such as the David-Bouldin index, leads to a smaller number of clusters while ensuring 95% usage coverage by the test suite. Several approaches from the state of the art are combined with UCDC and compared, and two of them outperform the others.