Quality Evaluation of Clustering-Based Process Model Discovery Using Event Log Similarity

Abstract

Model discovery discovers process models from event logs, but directly discovered models are usually complex and hard to interpret. Trace clustering tackles this problem by splitting an event log into smaller sublogs, each of which can support the discovery of simpler models. In this setting, clustering quality means that the obtained sublogs are behaviorally coherent and sufficiently different to achieve meaningful process model discovery. Existing evaluation depends on F-Measure, but it is dependent on the chosen discovery algorithm, requires costly fitness and precision computation, etc. To overcome these disadvantages, in this paper, we propose a clustering evaluation approach as a surrogate complement to model-quality evaluation rather than an alternative to existing model-quality measures such as fitness, precision, and F-Measure. Specifically, this clustering evaluation approach assesses clustering quality on the inter-sublog event log similarity, rather than the discovered process models. These are instantiated with three state-of-the-art log-similarity metrics. Our experiments using four real-life event logs show that the obtained scores are strongly negatively correlated with the F-Measure in most of the cases and have reduced the evaluation time by a factor of 47,547 down to 63 ms. These results indicate that the proposed framework provides an efficient and discovery-relevant alternative for clustering evaluation in process model discovery.



Received: 3 February 2026 | Revised: 10 July 2026 | Accepted: 16 August 2026



Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.



Data Availability Statement

The data that support the findings of this study are openly available in 4TU.ResearchData at https://doi.org/10.4121/uuid:aa01a720-4616-43e9-af67-370942019f48, https://doi.org/10.4121/uuid:60383406-ffcd-441f-aa5e 4ec763426b76, https://doi.org/10.4121/uuid:3926db30-f712-4394-aebc-75976070e91f, and https://doi.org/10.4121/uuid:ed445cdd-27d5-4d77-a1f7-59fe7360cfbe.



Author Contribution Statement

Guodan Yin: Data curation, Writing – original draft. Huiling Li: Formal analysis, Writing – original draft, Visualization. Cong Liu: Software, Validation, Writing – review & editing, Project administration, Funding acquisition. Qinghua Zhang: Methodology, Data curation, Investigation. Haijun Zhang: Conceptualization, Resources, Supervision.