| Software testing is an important phase to guarantee the quality of software. Debugging is a laborious job mainly by artificial means, so that fault localization becomes a focus field. In recent years, the spectrum-based fault localization techniques have attracted more and more research interest. The program spectra are collected by instrumentation in programs before test cases running. Then these techniques take this kind of information to correlate program entities and failed test cases so that the faulty entities can be found. However, there is a lot to be improved. So, the research on software fault localization is useful for increase efficiency in software developing, and it helps to free developers from tedious debugging work. The contribution of this paper includes:1. Review the existing spectrum-based fault localization techniques, and define the program spectra and the fault localization problem more formally.2. Propose a similarity coefficient fault localization technique, named Multi-Ochiai. This method uses the candidate of faults distribution to describe the location of multiple bugs in program. Genetic algorithm is employed to search for candidates with high suspiciousness values, and its results are converted to statements suspiciousness ranking.3. Propose a fault localization based on support vector machine. Program spectra and test case results are used to train a support vector machine, which will classify vast simulative test cases. Group voting method is used to get the suspiciousness ranking.4. An experiment is conducted to compare the two new methods with other three methods, and the Friedman test and multiple compare test are used to investigate the statistical significance of any differences observed in the experiments. The result suggests that the new methods outperform other related techniques in some respects. |
