Prefilter: A Fault Localization Method using Unlabelled Test Cases based on K-Means Clustering and Similarity

Abstract

Current research begins to apply unlabelled test cases to fault localization. However, in these methods, the labeled test cases randomly selected as the basis for fault localization cannot cover enough execution information, which will reduce fault localization efficiency. In this paper, a method based on K-Means clustering and similarity is proposed. At the beginning of the test, K-Means clustering is performed on the test case suite and the test cases filtered can cover more execution information. Next, for the test cases with failed execution results, the test cases with similar execution information are filtered to better highlight the fault information in the failed test cases. Experiments on Defects4J datasets show that the proposed method can be combined with other technologies to improve their efficiency, and the proposed method also has good compatibility with traditional software fault localization algorithms. The average improvement reached 13.37% in 8 scenarios.