Clustering of Usage Traces for Regression Test Cases Selection

Our objective is to define a regression testing approach that relies on usage traces that capture the behaviours of the system when exploited by the users. We achieve that by studying and evaluating clustering techniques applied to usage traces for regression tests selection. We first evaluate the existing vectorization methods and the clusters computed by the classical algorithms, and then, evaluate the clusters using existing state-of-the-art validation methods. We conclude that neither the existing clustering algorithms, nor the seminal clustering evaluation techniques are well-suited for identifying representative behaviours of the system when applied to usage traces. Thus, we propose a custom clustering algorithm and a dedicated cluster evaluation index for selecting usage trace to be used as regression tests.