A Fault Localization Method Based on Metrics Combination

Abstract

Spectrum-Based Fault Localization (SBFL) is one of the most effective fault localization techniques, and its performance closely depends on the program spectra and the ranking formula. Despite the numerous proposed approaches for fault localization, there are still great demands for fault localization techniques that can help guide developers to the locations of faults. Therefore, this paper defines four metrics from the program spectrum, which can become essential components of ranking formulas to mitigate spectrum-based fault localization problems. These metrics are further combined to propose a new heuristic, Metrics Combination (MECO), which does not require any prior information on program structure or semantics to locate faults effectively. The evaluation experiments are conducted on the Defects4J and SIR datasets, and MECO is compared with the 18 maximal formulas. The experimental result shows that MECO is more efficient in terms of Precision, Accuracy, and Wasted Efforts than the compared formulas. An empirical evaluation also indicates that two of the defined metrics, Assumption Proportion and Fault Assumption, when combined with the existing formulas, improve the localization effectiveness, especially the precision of ER5a-c (77.77%), GP02 (41%), and GP19 (27.22%), respectively.