Predicting higher order mutation score based on machine learning

Abstract

In software testing, the quality of the test suite plays a very important role for not only the effectiveness of the testing but also the quality assurance of software. Mutation testing is considered as the usable, automatic and very effective technique in detecting mistakes of the set of test cases such as missing test cases, redundant test cases··· However, when using the mutation testing technique in practice, the generation of a large number of mutants has led to very high computational costs. This raises the question of whether we can reliably and accurately predict this mutation score without running mutants or not. If we can do this, it will save a lot of time and effort but still ensure the effectiveness of mutation testing. In this paper, we propose the approach using machine learning to perform mutation score cross-prediction for software which are new and completely different from the software used to generate test data (mutants) in model training and testing. The experimental results have shown that our proposed approach has achieved the positive results and is highly feasible. Thus, we believe that the approach can be applied to significantly reduce the cost of mutation testing.