Pre-Processing Inertial Measurement Unit-based Data for Process Mining using Convolutional Neural Networks

Authors

Daniel Pollee, Department of Business Informatics and Process Analytics, University of Bayreuth, GermanyFollow
Milda Aleknonyté-Resch, Department of Computer Science, Kiel University, GermanyFollow
Dominik Janssen, Department of Business Informatics and Process Analytics, University of Bayreuth, GermanyFollow
Clint Hansen, Department of Neurology, Kiel University, GermanyFollow
Elke Warmerdam, Werner Siemens Endowed Chair of Innovative Implant Development (Fracture Healing), Saarland University, GermanyFollow
Walter Maetzler, Department of Neurology, Kiel University, GermanyFollow
Agnes Koschmider, Department of Business Informatics and Process Analytics, University of Bayreuth, GermanyFollow

Abstract

The analysis of inertial measurement unit (IMU)-based data allows tracking human behavior, detecting anomalies, and predicting human activity changes. As IMU-based data is unstructured and continuous, the application of process mining could provide additional insights into the underlying human performance. Therefore, the data has to be efficiently pre-processed in order to be used by process mining algorithms. This paper presents an approach to convert IMU-based data into structured event data for process mining. Particularly, the approach relies on methods for time-series segmentation and convolutional neural networks. In this way, activities of daily living can be identified from the unstructured data. The evaluation results show that convolutional neural networks are suitable for discovering activities when window sizes are previously known and have low cutoff values. The combination with a fixed sliding window approach for unknown window sizes appears superior.

Recommended Citation

Pollee, Daniel; Aleknonyté-Resch, Milda; Janssen, Dominik; Hansen, Clint; Warmerdam, Elke; Maetzler, Walter; and Koschmider, Agnes, "Pre-Processing Inertial Measurement Unit-based Data for Process Mining using Convolutional Neural Networks" (2024). Wirtschaftsinformatik 2024 Proceedings. 68.
https://aisel.aisnet.org/wi2024/68