Predictive business process monitoring with Markov models and reinforcement learning

Abstract

Informed decision making is at the forefront of many disciplines. Despite the domain we can always make preferable and educated decisions. This can lead to more insight, formulation of new and improved policies, always based on quality data. Specifically in the business domain, an organization can have more consistent results, transparency throughout the whole business process, improved performance and increased revenue. With the usage of process mining techniques, behaviors of processes can be analyzed and improved. Also modeling uncertainty through the usage of stochastic models can be an additional factor in the attempt to either or both improve and understand existing process and behaviors. The trend of automatic core functions in complex system such as an organization is even more relevant with the rapid advancement of artificial intelligence and more specifically machine leaning. A common stochastic model is Markov decision process (MDP) and MDPs are used to model reinforcement learning environments. With the goal of combining process mining, stochastic models particularly MDPs and reinforcement learning we aim to answer the question. In this thesis we strive to answer the question: Can predictive process monitoring be automated and if so at what level. Using a Business Process Intelligence challenge dataset and more specifically the 2017 Offer event log dataset. We attempt to further research the usage and application of these tools in the monitoring of business processes.