Got Flow?: Using Machine Learning on Physiological Data to Classify Flow

Raphael Rissler, Mario Nadj, Maximilian Xiling Li, Michael Thomas Knierim, Alexander Maedche

Abstract: As information technologies (IT) are both, drivers of highly engaging experiences and sources of disruptions at work, the phenomenon of flow - defined as "the holistic sensation that people feel when they act with total involvement" [5, p. 36] - has been suggested as promising vehicle to understand and enhance user behavior. Despite the growing relevance of flow at work, contemporary measurement approaches of flow are of subjective and retrospective nature, limiting our possibilities to investigate and support flow in a reliable and timely manner. Hence, we require objective and real-time classification of flow. To address this issue, this article combines recent theoretical considerations from psychology and experimental research on the physiology of flow with machine learning (ML). The overall aim is to build classifiers to distinguish flow states (i.e., low and high flow). Our results indicate that flow-classifiers can be derived from physiological signals. Cardiac features seem to play an important role in this process resulting in an accuracy of 72.3%. Hereby, our findings may serve as foundation for future work aiming to build flow-aware IT-systems.