Trace Clustering Exploration for Detecting Sudden Drift: A Case Study in Logistic Process

Abstract

Handling concept drift in process mining is one of the challenges tasks to construct the process model. Process model discovery, as the crucial perspective of process mining, should consider concept drift to discover the changes of a business process over time from execution trace. Many previous works have been dedicated to detect the drift using approaches from process discovery. As a matter of fact, there are many statistical parameters involving in the existing approaches that could be a barrier to construct the representative model. Unsupervised learning (i.e., trace clustering in process mining) could be the option to understand the changes of a process through learning the sequential patterns. However, there was a limited study on using trace clustering to detect the concept drift. This study attempts to explore the use of trace clustering techniques (i.e., profiles) to deal with change process discovery as the one of categorization of concept drift in process mining. The results of various trace clustering approaches were compared to the ground truth determined by both domain experts and existing concept drift approach. To verify the results, a dataset from logistics process was used. The case study in logistics process shows that partition-based clustering could be used to understand the concept drift.