A Fault Localization Method Based on Similarity Weighting with Unlabeled Test Cases

Abstract

In software fault localization problems, existing fault localization algorithms usually rely heavily on the perfection of test oracle. But in practice, there are a large number of test cases that lack accurate execution results. In order to utilize on unlabeled test cases, many test prediction and use case filter methods have been proposed. However, these methods ignore the similarity between test cases, which has been proven effective in fault localization studies using labeled test cases. Therefore, this paper proposes a fault localization method based on similarity weighting with unlabeled test cases. It uses the similarity of unlabeled test cases filtered by information entropy and labeled failed test cases as weights, and weights the suspicion calculation coefficients to enhance the importance of use cases similar to the failed cases. The experimental results show that similarity weighting effectively improves fault localization efficiency on all three program sets and all three localization algorithms. It can be seen that similarity of use case information also has an important role in the use of unlabeled test cases.