Spectrum‐based fault localization using empirical mode decomposition algorithm

Abstract

Spectrum‐based fault localization (SBFL) is considered as the most popular lightweight fault localization method. However, pure SBFL is proved to be tedious and time‐consuming for programmers to detect faults. This is because the suspiciousness is duplicated and they usually involve only the first few suspicious elements in the debugging process before losing patience. For this reason, benefiting from abundant spectrum information created by SBFL, we propose a new spectral fault localization technique using empirical mode decomposition method (EMD) to improve the accuracy of automatic software debugging. To accomplish that, the faulty program is evaluated by SBFL, and then, the resultant suspiciousness scores are taken as signals and nonfaulty elements as noise. Next, EMD is employed to decompose the suspicious scores of SBFL to eliminate massively repeated ones. Hence, elements are reranked according to new scores, and the ranking list is reconstructed by enlarging the high‐performance range (checking 5% elements) and TOP‐5 region to detect more faults. The performance of EMD‐SBFL is tested and compared with pure SBFL with EXAM scores and Top‐n ranks both on Siemens programs with seeded faults and large‐sized Defects4j programs with real faults. The result reveals that EMD‐SBFL is significantly effective to locate nearly over doubled faults by only check Top‐1 and outperforms the state‐of‐the art SBFL techniques.