Extracting the Representative Failure Executions via Clustering Analysis Based on Markov Profile Model

• Published in: Advanced Data Mining and Applications pp. 217 - 224
• Main Authors: Mao, Chengying, Lu, Yansheng
• Format: Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg 2005
• Series: Lecture Notes in Computer Science
• Subjects: Control Flow Graph; Prioritize Test Case; Profile Matrix; Subject Program; Test Execution
• ISBN: 354027894X; 9783540278948
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/11527503_26

During the debugging of a program to be released, it is unnecessary and impractical for developers to check every failure execution. How to extract the typical ones from the vast set of failure executions is very important for reducing the debugging efforts. In this paper, a revised Markov model used to depict program behaviors is presented firstly. Based on this model, the dissimilarity of two profile matrixes is also defined. After separating the failure executions and non-failure executions into two different subsets, iterative partition clustering and a sampling strategy called priority-ranked n-per-cluster are employed to extract representative failure executions. Finally, with the assistance of our prototype CppTest, we have performed experiment on five subject programs. The results show that the clustering and sampling techniques based on revised Markov model is more effective to find faults than Podgurski’s method.