On Clustering of Traces by Behavioral Similarity

Clustering approaches provide one solution to structure log
data and lay the foundation to efficiently analyze traces, which are further processed by process mining algorithms. In the context of
flexible processes, clustering approaches are a promising solution to mine more structured and simpler process models instead of spaghetti models, which are harmful with respect to analysis and understandability. To advance the field of event log clustering, this paper presents a novel approach to structure event logs. Existing clustering approaches take historical traces as input and subsequently intend to find a group of similar traces.
The clustering approach presented in this paper first constructs a classification and then assigns traces to related groups according to their behavioral similarity. In this way, clustering for both historical and realtime traces is possible. Beyond the theory, a visualization component was implemented allowing to detect homogeneous traces and to identify changes in traces related to time or any other exogenous factor.