Software Failure Detection Using Pattern´s Position Distribution

Pattern-based software failure detection is an important topic of research in recent years. In this method, a set of patterns from program execution traces are extracted, and represented as features, while their occurrence frequencies are treated as the corresponding feature values. But this conventional method has its limitation due to neglect the pattern´s position information, which is important for the classification of program traces. Patterns occurs in the different positions of the trace are likely to represent different meanings. In this paper, we present a novel approach for using pattern´s position distribution as features to detect software failure. In this method, we divide sequence into several sections and then compute pattern´s distribution in each section, the distribution of all patterns are then used as features to train a classifier. This method outperforms conventional frequency based method due to effectively identify software failures occur through mis-used software patterns. The comparative experiments in both artificial and real datasets show the effectiveness of this method.