Subgraph Mining for Anomalous Pattern Discovery in Event Logs

Conformance checking allows organizations to verify whether their IT system complies with the prescribed behavior by comparing process executions recorded by the IT system against a process model (representing the normative behavior). However, most of the existing techniques are only able to identify low-level deviations, which provide a scarce support to investigate what actually happened when a process execution deviates from the specification. In this work, we introduce an approach to extract recurrent deviations from historical logging data and generate anomalous patterns representing high-level deviations. These patterns provide analysts with a valuable aid for investigating nonconforming behaviors; moreover, they can be exploited to detect high-level deviations during conformance checking. To identify anomalous behaviors from historical logging data, we apply frequent subgraph mining techniques together with an ad-hoc conformance checking technique. Anomalous patterns are then derived by applying frequent items algorithms to determine highly-correlated deviations, among which ordering relations are inferred. The approach has been validated by means of a set of experiments.