Graph-based clustering for detecting frequent patterns in event log data

Finding frequent patterns is an important problem in data mining. We have devised a method for detecting frequent patterns in event log data. By representing events in a graph structure, we can generate clusters of frequently co-occurring events. This method is compared with basic association mining techniques and found to give a “macro-level” overview of patterns, which is more interpretable. In addition, the resulting graph-based clustering output for frequently co-occurring event sets is substantially less than association mining, while providing similar information levels. Therefore, the results are more manageable for practical applications.