Bayesian Network Based Predictions of Business Processes

Stephen Pauwels, Toon Calders

Abstract: Predicting the next event(s) in Business Processes is becoming more important as more and more systems are getting automated. Predicting deviating behaviour early on in a process can ensure that possible errors are identified and corrected or that unwanted delays are avoided. We propose to use Bayesian Networks to capture dependencies between the attributes in a log to obtain a fine-grained prediction of the next activity. Elaborate comparisons show that our model performs at par with the state-of-the-art methods. Our model, however, has the additional benefit of explainability; due to its underlying Bayesian Network, it is capable of providing a comprehensible explanation of why a prediction is made. Furthermore, the runtimes of our learning algorithm are orders of magnitude lower than those state-of-the-art methods that are based on deep neural networks.