Improving test suites via a novel testing with model learning approach

The integrity and reliability of reactive systems are very important and, in order to ensure them, an intensive testing procedure is required. This paper is concerned with the challenge of automatically producing test sets for black-box systems and proposes a testing with model learning technique to achieve functional coverage in the absence of specifications. The testing technique probes the system behaviour with tests, uses the test results to learn a behavioural model of the SUT, generates further tests on the learned model and refines it via inductive learning. The proposed inductive learning algorithm is based on Evidence Driven State Merging and introduces a novel heuristic approach according to the order in which state pairs are chosen for merging that drastically reduces the number of required merge operations. The effectiveness of the proposed testing technique is measured in terms of achieved functional coverage and test depth considering as a case study the event-based functional testing of an Android Application through its Graphical User Interface (GUI).