Behaviour Detection and Analysis in Process Mining - Discover Context Activities, Concept Drifts and Switch Behaviours in Event Logs

Abstract

This thesis undertakes an in-depth exploration of process mining, specifically focusing on the detection and analysis of three distinct types of process behaviours: context activities, process drifts, and switch behaviours. The first part of the thesis addresses the issue of context activities—activities that do not strictly follow the system's control flow. A novel method is proposed for their automatic detection from event logs, thereby improving the accuracy of discovered process models and distinguishing between context activities and outliers. The second part shifts the focus to process drifts, which are changes in business processes over time due to various factors. The contribution here lies in the development of a robust method for detecting such drifts, particularly in event logs with noise or infrequent behaviours. This is achieved through statistical validation, ensuring that identified drifts are not mere fluctuations but statistically significant changes. Additionally, a method for detecting changes in branching frequencies, a specific type of process drift, is introduced. Finally, the thesis enhances existing process discovery algorithms, with a particular focus on improving the capabilities of the Inductive Miner. A novel extension to the traditional process tree model is proposed to accommodate switch behaviours, thereby enhancing the algorithm's ability to produce more accurate and flexible process models. Collectively, these contributions aim to improve the robustness and flexibility of process mining techniques, making the analysis of event logs more reflective of real-world complexities.