FrMi: Fault-revealing Mutant Identification using killability severity

Context: Mutation testing is a powerful method used in software testing for various activities, such as guidance for test case generation and test suite quality assessment. However, a vast number of mutants, most unrelated to real faults, threaten the scalability and validity of the method. Over the decades, researchers have proposed various approaches to alleviate these problems, most of which have almost the same performance in practice. To overcome this issue, recently predicting a category of mutants named fault-revealing mutants has been proposed, which outperforms other methods in terms of real-fault revelation ability. Although recent research shows the usefulness of targeting this type of mutant, they are scarce, which makes predictions of them with higher accuracy challengingObjective: This paper aims to propose a method that can predict fault-revealing mutants with higher accuracy compared to the state-of-the-art method.Methods: To tackle this challenge, a feature representing the difficulty of killing a mutant is added as a new feature to complement the state-of-the-art feature set. Then a method based on ensemble learning is proposed that uses this feature for fault-revealing mutants' prediction.Results: According to our experimental results, the proposed method outperforms the state-of-the-art method regarding area under a receiver operating characteristic curve (AUC) value on the Codeflaws and CoRBench data sets by 7.09% and 8.97%, respectively. Conclusion: It is concluded that the proposed method, which includes a new feature and an ensemble-learning approach, enhances the accuracy of predicting fault-revealing mutants in software testing. This is achieved by incorporating the difficulty of killing a mutant as a feature, which complements the existing feature set used in state-of-the-art methods. The experimental results demonstrate that the proposed method outperforms the state-of-the-art method on two datasets, Codeflaws and CoRBench, indicating that it has the potential to be applied in practical software testing scenarios.