Recognizing chaotic activity during the event data pre-processing phase

Abstract

By extracting knowledge from event data, process mining tries to get insights into corporate processes. Indeed, the quality of events is an important factor in developing process models that match business process reality. To that end, pre-processing approaches have emerged to cleanse events of defects (noise, incompleteness, and rare behaviours) at the limit of chaotic activity formation. In the process, chaotic actions are carried out randomly, affecting the quality of found models. In addition, a supervised learning strategy employing labelled samples to detect chaotic behaviours has been presented. This highlights the difficulties of characterising chaotic activities in the absence of prior knowledge about which activities are actually chaotic. To that aim, we design a method for recognising chaotic events in the absence of labelled training data.