Using Pattern Position Distribution for Software Failure Detection

In this paper, we present a novel approach to software failure detection based on pattern position distributions as features. In this approach, we divide an execution sequence into several sections and then compute a pattern distribution in each section. The distribution of all patterns is then used as a feature to train a classifier. This approach outperforms conventional frequency based methods by more effectively identifying software failures occurring through misused software patterns. Comparative experiments show the effectiveness of our approach.