Automated discovery of business process simulation models from event logs: a hybrid process mining and deep learning approach

Doctoral theses defended at the University of Tartu, summary in Estonian.

For companies, changing a process can be costly and risky but necessary. And not doing it can affect its resources, its environment, or its continuity. One of the techniques most used by companies to design and evaluate their processes is business process simulation. This technique allows creating hypothetical scenarios and assessing the consequences of their implementation in a virtual environment without taking the risks of failure in real life. Modifying the process components in the simulator allows the analysts to make assumptions such as "if you remove this, this could happen, or if you add this, this could happen." This ability is very convenient to support the decision-making process concerning potential changes. The problem with this method is that creating and fitting a simulation model is a complex task that requires time and specialized technical knowledge. In addition, analysts usually create simulation models through interviews, observations, and sampling. All these techniques are highly prone to biases, which means that the precision of these manually created models is relatively inaccurate. All this makes disappointing the adoption of business process simulation, making it difficult for companies to use this technique. This thesis proposes new techniques for creating business process simulation models that use data extracted from enterprise information systems in conjunction with neural networks and process mining algorithms. This thesis aims to create a more precise automatic simulation technique that requires less human intervention solving the drawbacks of the current process simulation approach. We consolidate the techniques proposed in this thesis in two open-source tools. The first tool, called Simod, can fully automatically discover and fine-tune simulation models through process mining techniques. However, the proposed method often falls short when it comes to predicting the timing of each activity. In response, the second tool called DeepSimulator combines discovery techniques based on process mining with generative models based on deep learning. The evaluation results of this hybrid approach lead to simulations that more closely reflect the observed dynamics of the process than methods based purely on process mining or deep learning.