that are closest to one another and merge them. This is repeated until all instances are in the same cluster. Figure 3.7(a) shows all intermediate clusters, i.e., all except the initial singleton clusters and the final overall cluster. Because of the hierarchical nature of the agglomerative hierarchical clustering, we can visualize the clusters using a so-called dendrogram as shown in Fig. 3.7(b).
Any horizontal line cutting through the dendrogram corresponds to a concrete clustering. For example, Fig. 3.8(b) shows such a horizontal line. The clusters resulting from this are shown in Fig. 3.8(a). Moving the line to the bottom of the dendrogram results in many singleton clusters. Moving the line all the way up results in a single cluster containing all instances. By moving the horizontal line, the user can vary the abstraction level.
Clustering is only indirectly related to process discovery as described in Chap. 1. Nevertheless, clustering can be used as a preprocessing step for process mining [12, 32, 46]. By grouping similar cases together it may be possible to construct partial process models that are easier to understand. If the process model discovered for all cases is too complex to comprehend, then it may be useful to first identify clusters and then discover simpler models per cluster.