Information entropy-based clustering algorithm for rapid software fault diagnosis

In order to rapidly diagnose and locate the fault, we present ICARSFD (Information entropy-based Clustering Algorithm for Rapid Software Fault Diagnosis). In this paper, the average entropy and the total entropy are defined to guide the clustering operation over fault modes. This algorithm firstly stores the related information of existing faults in the form of fault tree, and deems each fault as an initial cluster. By calculating the information entropy between clusters and comparing them with the average entropy and the total entropy, fault clustering is completed. For the faults inappropriate to their located clusters, we take a retrospective approach to cluster them. Thereby the clustering effect related with the fault order could be addressed. Secondly, according to the ascending order of information entropy, the fault to be analyzed is matched to each cluster. Lastly, both the fault diagnosis results and fault paths are put out. In addition, if the fault match isn't successful, the fault path will be identified through fault tree, and the clustering results will be updated later. The experimental results demonstrate that ICARSFD has both good clustering effect and detection effect.