Multi-objective Search for Model-based Testing
Rui Wang, Cyrille Valentin Artho, Lars Michael Kristensen, Volker Stolz · 2020
This paper presents a search-based approach relying on multi-objective reinforcement learning and optimization for test case generation in model-based software testing. Our approach considers test case generation as an exploration versus exploitation dilemma, and we address this dilemma by implementing a particular strategy of multi-objective multi-armed bandits with multiple rewards. After optimizing our strategy using the jMetal multi-objective optimization framework, the resulting parameter setting is then used by an extended version of the Modbat tool for model-based testing. We experimentally evaluate our search-based approach on a collection of examples, such as the ZooKeeper distributed service and PostgreSQL database system, by comparing it to the use of random search for test case generation. Our results show that test cases generated using our search-based approach can obtain more predictable and better state/transition coverage, find failures earlier, and provide improved path coverage.