Fault localization based on non-parametric hypothesis testing model

The size of failed runs and the distributions of the evaluation biases on the Siemens suite were investigated. The empirical results show that the precondition for the central limit theorem and the assumption on feature spectra forming normal distributions are not well-supported by empirical data. Thus, the previous method based on parametric hypothesis testing has a potential problem. New approaches based on non-parametric hypothesis testing models were proposed. The empirical results on the Siemens suite indicate that these approaches can outperform existing predicate-based statistical fault localization techniques, especially on non-normal distributions. The effectiveness comparison between two methods based on two popular non-parametric hypothesis testing models was also investigated. One method was based on the Kolmogorov-Smirnov test and the other was based on Cram閞-von-Mises criterion. The comparison results show that the method based on the Kolmogorov-Smirnov test consistently outperforms that based on the Cram閞-von-Mises criterion in the task of fault localization.