Generating Transition Probabilities for Automatic Model-Based Test Generation

Abstract

Markov chains with Labelled Transitions can be used to generate test cases in a model-based approach. These test cases are generated by random walks on the model according to probabilities associated with transitions. When these probabilities correspond to a usage profile, reliability may be estimated. However, in early stages of development, such probabilities are not easy to determine, thus default profiles must be considered. In such a case it may be interesting to target some coverage criteria rather to use classical uniform probability generation approach. In this paper we enrich an existing industrial tool based on usage profile with 3 possibilities to create default profiles that improve transition coverage. We report experiments that compare the improvement of the coverage rates by our approaches with respect to uniform probabilities on transitions from a given state, which is the current default profile.