Performance Analysis: Discovering Semi-Markov Models From Event Logs

Process mining methods and tools are largely used in industry to monitor and improve operational processes. This paper presents a new technique to analyze performance characteristics of processes using event data. Based on event sequences and their timestamps, semi-Markov models are discovered. The discovered models are further used for performance what-if analysis of the processes. The paper studies a trade-off between the order of models discovered and accuracy of representing performance information. The proposed discovery and analysis techniques are implemented and tested on real-world event data.