Hyper-Heuristic Based Product Selection for Software Product Line Testing

The Feature Model (FM) is a fundamental artifact of the Software Product Line (SPL) engineering. It represents all SPL variabilities and commonalities and is generally used to derive products for SPL testing. However, the testing of all products is almost impossible in practice. Therefore, search-based approaches can be found in the literature to select the most interesting ones. Among them, the approaches that use Multi-Objective Evolutionary Algorithms (MOEAs) are very promising since this selection problem is impacted by many factors. Yet for the tester it is not always easy to choose the best algorithm, and to configure parameters and operators to solve the problem. To help in this task, we introduce in this paper a Hyper-Heuristic (HH) approach. HHs are methodologies used to select or generate heuristics. Our HH approach dynamically selects the best operators, considering four objectives of the problem: the number of products, pairwise coverage, mutation score, and dissimilarity of products. The approach is implemented and evaluated with four MOEAs: NSGA-II, SPEA2, IBEA, and MOEA/D-DRA, and two selection methods: random and Upper Confidence Bound (UCB) based. Evaluation results show that the HH-based NSGA-II algorithm generates the best results. Moreover, the UCB method outperforms the random selection on large instances.