Process-level Model Repair through Instance Graph Representation

Existing model repair techniques propose changes that are based on event-level deviations observed in a log, like inserted or skipped events, often overlooking process precision at the advantage of fitness. The present short paper aims to briefly introduce the recent proposal of an alternative approach targeting higher-level structured anomalous behaviors. To do this, the approach exploits instance graph representation of anomalous behaviors, that can be derived from the event log and the original process model. The approach demonstrates that repaired models obtained in this way show higher precision and simplicity, with only small reduction of process fitness.