Answers are grounded in this paper and cite their sources.
Software fault localization is a crucial technique for effectively identifying faulty statements. While several methods have been proposed for handling single faults, real‐world programs often exhibit multiple fault behaviors. Past studies on multiple fault localizations (MFL) have explored ensemble methods, clustering, and more. However, larger program sizes lead to increased execution time, overhead, and resource consumption. To address these issues, we introduce two‐phase clustering technique. In the initial phase, we leverage coverage data to reduce the number of failed test cases and executable statements. This phase also enables early prediction of both single and multiple faults for failed test cases. In the subsequent phase, we employ single linkage hierarchical clustering with three different ensemble‐weighted distances and implement two‐round fault localizations with SBFL or deep learning‐based methods to pinpoint faulty statement locations. To validate the performance, we conducted experiments using the three well‐known open‐source Linux utility programs gzip, grep and sed, totaling 149 different versions with multiple faults. The experimental results show that the AVG (T' EXAM score) for two‐phase clustering decreased from 16.933% to 7.637%, while MSeer decreased from 21.086% to 8.779%, and the cluster‐free decreased from 37.155% to 20.629%. The total cumulative statements examined in two‐phase clustering improved by 32.575% compared to MSeer and, by 66.701% compared to cluster‐free.
No takes yet. Share an insight, caveat, or question.
Fang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: