A Trace Clustering Framework for Improving the Behavioral and Structural Quality of Process Models in Process Mining

Abstract

Process mining (PM) techniques are increasingly used to enhance operational procedures. However, applying PM to unstructured processes can result in complex process models that are difficult to interpret. Trace clustering is the most prevalent method for handling this complexity, but it has limitations in dealing with event logs that contain many activities with varied behaviours. In such cases, trace clustering can produce inaccurate process models that are expensive in terms of time performance. Therefore, it is crucial to develop a trace clustering solution that is optimal in terms of behavioural and structural quality of process models while being efficient in terms of time performance. In this study, we introduce a refined trace clustering framework with an integration of log abstraction and decomposition technique that improves the precision of process models by 38%, leading to a 40% increase in the f-score. The proposed framework also produces process models that are 38% simpler than those produced by baseline approaches. More importantly, our framework achieves a remarkable 89% improvement in time performance, making it a valuable contribution to the field of process mining. Future works include exploring the scalability of the proposed framework against a wider range of complex event logs and testing the framework to validate its effectiveness in practical applications.