Abstract
Fault localization (FL) is an essential phase in software debugging and is used to detect the possible faulty location. Nowadays, deep learning technique has a great advantage to extract semantic features of program and hidden data information for coverage data effectively. Therefore, deep learning-based FL is regarded as the latest solution. However, it is well known that none of existing methods is suitable to all scenarios including single fault and multiple faults. Furthermore, the optimal hyperparameters of deep learning model for FL are hard to be determined in different types of programs. According to past researches, Multicriteria decision-Making (MCDM) can provide the best or trade-off solutions to multiple method combinations. In this paper, we combined multiple-criteria ensemble weight strategy with deep learning-based FL to improve the effectiveness of program. Overall, our proposed method has also a good scalability and run some experimental results to obtain significant performance improvement comparisons among the past methods.