A research prototype for process model analysis and reference model mining. This webseite is currently under construction. Thank you for your patience.
Run RefMod-Miner as a Service now!

Model clustering



Basic concept:

Nowadays, companies are working with an ever-increasing amount of Business processes which must be regularly evaluated for quality assurance.
The basis for this evaluation is provided by through process-supporting business software (for example ERP) generated process logs.
The evaluation of these process logs by means of process mining and other analysis techniques are a highly complex task, since the underlying logs contain large amounts of heterogeneous information on fundamentally different processes and process variants.

This makes grouping, a „clustering“ of these elements into thematic coherent subgroups for the purpose of a meaningful analysis of the process protocols indispensable.

In clustering, data objects are generally divided in groups with similar properties. Basically, the methods of clustering are split into hierarchical, partioning and density-based processes.

Hierarchical methods connect objects with a small distance or with a high similarity together; Originating from the starting point, this method can be separated in divise (from a large clustering to many small ones) and agglomerative (small clusters to larger ones) algorithms.

In the case of partioning processes, first a specific number of desired clusters K is established. Subsequently, the data objects are transferred step by step between the clusters, until the cluster affiliation no longer changes.
Density-based methods consider the inner density of each cluster and set similarities based on the proximity of the individual points.


How the RefMod-Miner works:

The grouping in the RefMod-Miner takes place on the basis of specific similarity measures, that divide the models into thematically similar fields. For this operation, the user can currently choose from four different algorithms, using the different methods of clustering.

PAM (partioning around medrioids) and KMEANS are partioning procedures, HCLUST is a hierarchical device (the applied algorithm can be specified within the RefMod-Miner) and DBSCAN is a density-based method.

In addition to creating the clusters, the RefMod-Miner also supports its subsequent evaluation and further use.



Heatmap gruppiert nach Cluster