Research on ML-based software for BPMN database processing to predict CFC using TNN and TNSF

Abstract

Business process modelling is a widely used technique in enterprise management and information system development, with the quality of business process models determining the degree of their comprehensibility to users and the effectiveness of organizational activities. A business process model can be formally described using a directed labeled graph, which allows the application of complexity metrics, in particular the Control-Flow Complexity (CFC), for objective evaluation of model quality. This study proposes a novel approach to assessing the quality of BPMN (Business Process Model and Notation) models from the perspective of complexity, combining linear regression techniques for predicting CFC metric values based on structural characteristics and threshold values for classifying complexity levels. The developed linear regression model, trained on a set of 6137 BPMN descriptions, demonstrates high prediction quality with a coefficient of determination of 0.81 and a mean square error of 2.44. Experimental evaluation of the effectiveness of BPMN model classification by complexity levels showed the best results for low and high complexity levels with accuracy, precision, recall, and F1-measure scores ranging from 0.75 to 0.97, while the moderate level is characterized by lower accuracy due to its transitional nature.