Conformance Checking-based Concept Drift Detection in Process Mining

Abstract

One of the main challenges of process mining is to obtain models that represent a process as simply and accurately as possible. Both characteristics can be greatly influenced by changes in the control flow of the process throughout its life cycle. In this thesis we propose the use of conformance metrics to monitor such changes in a way that allows the division of the log into sub-logs representing different versions of the process over time. The validity of the hypothesis has been formally demonstrated, showing that all kinds of changes in the process flow can be captured using these approaches, including sudden, gradual drifts on both clean and noisy environments, where differentiating between anomalous executions and real changes can be tricky.