Hold me tight: Towards the detection of the flow experience using a commercial video game controller

The prolific consumption of video games has inspired a rich body of research to explain their ability to induce satisfying experiences. The flow experience is often defined as a state that encapsulates the essential qualities of engaging gameplay. However, there is limited research addressing the detection of flow, and existing approaches often involve intricate equipment requirements that compromise commercial applicability. In this study, we review the challenges behind flow's conceptualization and discuss how they leak into approaches that employ machine learning techniques. We propose a novel, multidimensional representation of flow that combines subjective and objective data – game controller interaction and player performance on a secondary oddball task – whilst treating flow detection as a regression problem. Contrary to previous studies, the game difficulty is divorced from the model's training, with the aim of improving generalizability across game titles. The results indicate high predictive accuracy with an average composite loss rate of 0.0681 (±0.0036) that significantly outperformed two baseline models. This finding suggests an effective mapping of the objective data onto self-reported flow and reinforces the viability of game controllers as a means for flow detection. In addition, we identified that the main drivers behind the model's predictions were of inertial origin. We conclude that these insights offer practical considerations for the future of interactive applications. • This study represents the flow state using an oddball task and self-reported flow. • Unlike previous studies, flow is not represented through game difficulty. • A transformer model learned to map flow using game controller interaction data. • The orientation and acceleration features had the highest contribution.