Predictive Analysis of Business Processes Using Neural Networks with Attention Mechanism

Abstract

The analysis of ongoing processes is an important task in business process management. This is not surprising, since (e.g.) being able to predict future events in processes enables companies to intervene at an early stage if deviations from a desired workflow are likely to occur. Subsequently, errors and associated financial losses can be prevented or avoided. A common basis for being able to predict future events is a sequence of previous events that are typically stored in a socalled event log. In this work we present a neural network with attention mechanism, which is trained using publicly available event logs (e.g. BPI Challenge 2013). Furthermore, we elaborate the proposed model with an extensive dataset of events of a worldwide known German software company. In addition to promising results (e.g.) with regard to n-gram models, the training time of the proposed model is shorter than that of typical reference models such as neural networks with a long short-term memory architecture (LSTM).