Machine Learning Driven Identification of Coincidental Correct Test Cases

Spectrum-Based Fault Localization (SBFL) technique has been extensively investigated recently. The primary objective of the SBFL technique is to expedite the identification of fault statements by developers with improved accuracy. However, test suites may contain Coincidental Correct (CC) test cases, where they execute the fault statement but still outputs the expected result. CC test cases may adversely affect the efficiency of fault localization. To enhance the precision of fault localization, we propose Identification of CC test cases using Machine Learning (ICCML) approach. ICCML approach extracts test-related features base on coverage information of the test suites and uses multiple SBFL formulas to expand the feature dimension. To measure the performance of ICCML, we performed experiments utilizing the Defects4J dataset. Compared with the baseline, ICCML approach achieves better CC test case identification accuracy. Furthermore, the fault localization effectiveness exceeds the baseline after using the Cleaning and Relabeling strategies.