Process Mining
In my master’s thesis with Prof. Matthias Weidlich at HU Berlin, I investigated trace profiles for predictive business process monitoring. Trace profiles map complex event log traces to a feature space that is suitable for machine learning algorithms. I applied existing and newly developed trace profiles to real-world datasets from the public administration (EU agricultural subsidy management) as part of my work with data experts.
In addition to diving deep into process discovery and preprocessing algorithms, I trained machine learning models to predict future (undesired) outcomes of currently running process instances. A major implementation challenge was integrating established process mining algorithms within the ProM framework with supervised (Random Forest) and unsupervised (One-class SVM) machine learning algorithms in the Java ecosystem and the Scala programming language.
As an important outcome of the project, we managed to share an updated, de-identified version of the event logs with the community as part of the International Business Process Intelligence Challenge 2018.
Links
- BPI Challenge 2018
- Master's thesis: Feature extraction from event logs for predictive monitoring of business processes
- Predictive Process Monitoring (GitHub)