ERP Event Log Preprocessing: Timestamps vs. Accounting Logic

Process mining has been gaining significant attention in academia and practice. A promising first step to apply process mining in the audit domain was taken with the mining of process instances from accounting data. However, the resulting process instances constitute graphs. Commonly, timestamp oriented event log formats require a sequential list of activities and do not support graph structures. Thus, event log based process mining techniques cannot readily be applied to accounting data. To close this gap, we present an algorithm that determines an activity sequence from accounting data. With this algorithm, mined process instance graphs can be decomposed in a way they fit into sequential event log formats. Event log based process mining techniques can then be used to construct process models. A case study demonstrates the effectiveness of the presented approach. Results reveal that the preprocessing of the event logs considerably improves the derived process models.