Complexity Clustering of BPMN Models: Initial Experiments with the K-means Algorithm

Abstract

This paper introduces a method to assess the complexity of process models by utilizing a cluster analysis technique. The presented method aims to facilitate multi-criteria decision making and process objective management, through the combination of specific quality indicators. This is achieved by leveraging established complexity metrics from literature, and combining three complementary ones (i.e. NOAJS, CFC and CNC) to a single weighted measure, offering an integrated scheme for evaluating complexity. K-means clustering algorithm is implemented on 87 eligible models, out of a repository of 1000 models, and classifies them to corollary clusters that correspond to complexity levels. By assigning weighted impact on specific complexity metrics -an action that leads to the production of threshold values- cluster centroids can fluctuate, thus produce customized model categorizations. The paper demonstrates the application of the proposed method on existing business process models from relevant literature. The assessment of their complexity is performed by comparing the weighted sum of each model to the defined thresholds and proves to be a straightforward and efficient procedure.