Master Thesis - Detecting surprising instances in a process



Mahnaz Qafari

Wissenschaftliche Mitarbeiterin


+49 241 80 21912




Faults and deficiencies cost companies millions or even billions each year and it is yet just one of their detriments. They also cause damages to the image of the companies in terms of reliability and accountability. That is why it is crucial for companies to constantly enhance their processes.

Process mining is a collection of approaches, techniques, and tools, which can be used to extract a wide range of knowledge about the processes from their event logs. One of the main purposes of applying process mining is process enhancement which starts with identifying a problem in the process and those cases that suffered from that problem. The next step is using machine learning or causal inference techniques to uncover the root cause of the problem.

As another source of information, we can use the cases with surprising results, i.e., those for which the process performs significantly better or worse than other similar cases. By investigating the root cause of such deviations in the behavior of the process, we can uncover not only the ground reasons for the poor behavior of the process, but also the conditions where the process performs much better. This knowledge can be turned into actionable steps to improve the process performance indicators.

In this thesis, we aim at defining what a surprising situation is, developing methods to discover them, and then turning this knowledge into actionable information. This involves not only the formal definition but also implementing the approaches and analyzing the results of applying it on the real and/or synthetic processes.

The ideal candidate has a strong background in process mining and is willing to perform experiments with the following techniques:

  • Machine learning techniques (clustering, classifying)
  • Graph-based techniques (random walk, graph machine learning).
  • Causality-based techniques

A multi-perspective assessment of the defined surprising situation (computation time, number of hyper-parameters, …) is needed. With this purpose, it is possible to run the experiments on the PADS HPC cluster or the RWTH compute cluster.

The assessment of the developed techniques will be done on several real-life event logs. After identifying the surprising situations and extracting the features, the application of machine learning techniques (classification, prediction, clustering) and causal inference techniques will contribute to turning the identified situations into actionable insight.


Process Mining Book


Python / Javascript programming

Process mining knowledge

Machine learning knowledge

Supervisor Wil van der Aalst


Mahnaz Sadat Qafari

For more Information

Send an e-mail to . Make sure to include detailed information about your background and scores for completed courses.