Mulr4FL: Effective Fault Localization of Evolution Software Based on Multivariate Logistic Regression Model

Fault localization is indeed tedious and costly work during software maintenance. Studies indicate that combining both structural features and behavior characteristics of programs can be beneficial for improving the efficiency of fault locating. In this paper, we proposed a framework, called Mulr4FL, for fault localization using a multivariate logistic regression model that combined both static and dynamic features collected from the program under debugging. Firstly, the hybrid metrics data set, with both program structural features and behavior characteristics combined, is constructed by static program analyzing and dynamically tracing that runs with a designed metrics set. Meanwhile, the fault information of the legacy program is also obtained from the bug tracking system. Secondly, Bivariate logistic analysis is performed to filter the metrics that are significantly related to faults, and then with the selected metrics and their measurements, a multivariate logistic regression model was constructed and trained. Finally, based on the trained logistic model, we conduct the multivariate logistic analysis on the features of the evolved software and predict the buggy class methods. An empirical study was conducted based on a set of benchmarks that are used widely in program debugging research. The results indicate that the Mulr4FL can significantly improve the effectiveness of locating faults in contrast to 5 baseline techniques.