A Framework for Unsupervised Event Abstraction using Hidden Markov Models
MSc Thesis Project
Student: Anand Kumaraguru
Title: A Framework for Unsupervised Event Abstraction using Hidden Markov Models
Supervisor: Dr. Sebastiaan J. van Zelst
1st Examiner: Prof. Wil van der Aalst
2nd Examinder: Prof. Horst Lichter
Information systems generate a wealth of data about the business processes being executed. Process Mining leverages this data to increase our understanding of the process by helping us model the underlying process, detect any discrepancies and overhead components slowing the execution. Under process mining, process discovery is a discipline that is concerned with unearthing the underlying model that describes the process. A process model built out of the real-life logs tend to be complex in nature and fails to capture the intrinsic structure of the process at the business level because of the difference in levels of granularity in which they are recorded. Event abstraction techniques in process mining attempt to address this problem by abstracting such data into the appropriate granularity level. In this thesis, we propose an unsupervised learning technique using hidden markov models that clusters the events together to produce a simplistic model. The proposed technique is then tested on real-life data sets. A comparison between the models discovered from the clustered group and the model discovered on the basis of the complete log is presented at the end.