MQP: Mutants Quality Prediction for Cost-Effective Mutation Testing
Xingya Wang, Shiyu Zhang, Fangxiao Liu, Lichao Feng, Zhihong Zhao · 2021 IEEE 21st International Conference on Software Quality, Reliability and Security Companion (QRS-C) · 2021
Mutation testing simulates typical software defects by executing mutation operations. It can effectively measure the defect detection capabilities of a test-suite. However, in the mutation testing, we must generate, compile, and execute plenty of mutants, which causes substantial computing costs. Mutants selection can reduce the number of mutants but cannot avoid compiling and executing all mutants. Therefore, mutation testing suffers an efficiency problem, making it hard to use in practice. In this paper, we propose a Mutants Quality Prediction method, MQP, which relies on the static features of the source program, test-suite, and mutants to predict the quality of a test-suite. MQP avoids compiling and executing all mutants during mutants selection and thus can save expensive computing costs. Then, we use MQP as the fitness function and select the evolutionary optimization strategy, Genetic Algorithm, to generate a high-quality mutant subset. We empirically studied MQP on nine widely used items. The experimental results show that MQP achieves a good predictive effect. Its accuracy rate can reach 98.30%, while the average absolute error is only 0.74%. Moreover, the reduction efficiency of the MQP-based mutation testing is about 20 times that of the random-based selective mutation testing.