Detecting concept drift in real-life event logs and using process variant analyses for examining the found drift

  • R.H.M. (Roel) Claessen

Student thesis: Master's Thesis

Abstract

We present an approach to detect concept drift in a real-life event-log and examine the found drift via variant analysis. The approach can help trace best practices in given time periods or departments. It can be utilized by anyone who wants to get a deeper understanding of their process and see where improvement is possible.
Unforeseen changes that happens over time in processes can occur through changing situations, for instance, amending regulation or seasonal effects. The phenomenon of unforeseen change over time is defined in the literature as concept drift.
There is research done on concept drift in the process mining community and solutions have been provided. We present a practical and business friendly approach to detect and examine concept drift, partly based on an existing solution. One solution in the form of a ProM plug-in is evaluated on a real-life event log which covers a considerable period. Furthermore, an unexpected applicability of the plug-in to demonstrate volatility in a process is presented. Moreover, we investigate how and to what extent found concept drift and volatility can be examined using variant analyses. Next to this we elaborate which shortcomings apply when using a real-life event log, especially when a profound business context is missing.
Date of Award21 Jun 2020
Original languageEnglish
SupervisorLloyd Rutledge (Examiner) & Stef Joosten (Co-assessor)

Keywords

  • Business Rule Mining
  • Process Mining
  • Concept Drift
  • Variant analyses
  • Volatility

Cite this

'