Using SOM for Extracting Process Model Relations from Event Logs

Abstract

The acquisition of knowledge plays a crucial role in improving business processes. Process discovery is a technique employed to obtain essential information about actual operations in the form of process models. However, the process discovery encounters various challenges that hinder the quality of the resulting process model. The misleading or low-quality process models are the results of the high degree of complexity of business processes as there are different ways in which activities can be ordered. Thus, this paper presents an automated approach for identifying the relationship between pairs of activities in a process model, a crucial aspect of the process discovery process. The method employs Self-Organizing Maps (SOM) from artificial neural networks for categorizing relationship types: direct-followed, parallelism, long-distance, and concurrent relationships. The model was trained using event logs generated from a simulation tool with varying degrees of complexity. The findings suggest that the proposed method exhibits outstanding results in categorizing some relationship types. Nonetheless, the accuracy of clustering for the nested pattern AND is not yet satisfactory.