An analysis of test case coverage balancing for fault localization

Context: Spectrum-based defect localization uses code coverage information obtained during testing to abstract patterns that help identify faulty statements. Since the number of test cases that reveal the presence of defects is typically lower than that which do not, such imbalance hinders the discovery of these patterns and, therefore, adds cost to defect localization. Objective: A better balanced test suite is necessary to achieve efficiency in software fault localization. This article explores the following questions: (1) Does balancing test cases reduce the effort involved in software fault localization? (2) If two or more test cases are redundant with respect to code coverage, does eliminating this redundancy improve the effectiveness of fault localization heuristics? Method: The Defects4J benchmark was used and the Smote technique (Synthetic Minority Over-sampling Technique) was applied to balance the test cases with and without removing redundant test cases with respect to code coverage. After balancing, the effectiveness of software fault localization was evaluated using some heuristics. Results: It was revealed that the removal of redundancy with respect to code coverage promoted the effectiveness of the fault localization heuristics when applied after the test case balancing procedure implemented by the Smote technique. Conclusions: The analyses presented in the article revealed the positive possibility of exploring research into the use of test case balancing techniques as a way of improving the effectiveness of fault localization heuristics.