Exploring the effect of context information on deep learning business process predictions

• Published in: Journal of decision systems Vol. 29; no. sup1; pp. 328 - 343
• Main Authors: Brunk, Jens, Stottmeister, Johannes, Weinzierl, Sven, Matzner, Martin, Becker, Jörg
• Format: Journal Article
• Language: English
• Published: Abingdon Taylor & Francis 18.08.2020; Taylor & Francis Ltd
• Subjects: Artificial neural networks; Business process management; Context; context-sensitivity; Deep learning; Machine learning; Neural networks; Prediction models; Predictions; Predictive process monitoring
• ISSN: 1246-0125; 2116-7052
• DOI: 10.1080/12460125.2020.1790183

Predictive Process Monitoring (PPM) techniques for predicting the next activity in running business processes developed into an established topic of Business Process Management. Recent research suggests using Deep Neural Networks (DNNs) for PPM because DNNs are good at learning the intricate structure of business processes. Most of these works use Long Short-Term Memory Neural Networks (LSTMs) and consider only the control flow information of an event log. Beyond control flow information, context information can add valuable information to a predictive model. However, the effects of context attributes on the predictive quality have not yet been sufficiently analyzed. This work addresses this gap and provides two insights. First, a context-sensitive prediction capability can improve the predictive quality of an LSTM-based technique. Second, the added value of context information to the quality of predicting the next activity varies in the course of a running process instance.