Most dynamic fault localization methods aim at totally ordering program elements from highly suspicious to innocent. This ignores the structure of the program and creates clusters of program elements where the relations between the elements are lost. We propose a data mining process that computes program element clusters and that also shows dependencies between program elements. Experimentations show that our process gives a comparable number of lines to analyze than the best related methods while providing a richer environment for the analysis. We also show that the method scales up by tuning the statistical indicators of the data mining process.
展开▼
