Predicting Video Game Players’ Fun from Physiological and Behavioural Data

Abstract

Finding a physiological signature of a player’s fun is a goal yet to be achieved in the field of adaptive gaming. The research presented in this paper tackles this issue by gathering physiological, behavioural and self-report data from over 200 participants who played off-the-shelf video games from the Assassin’s Creed series within a minimally invasive laboratory environment. By leveraging machine learning techniques the prediction of the player’s fun from its physiological and behavioural markers becomes a possibility. They provide clues as to which signals are the most relevant in establishing a physiological signature of the fun factor by providing an important score based on the predictive power of each signal. Identifying those markers and their impact will prove crucial in the development of adaptive video games. Adaptive games tailor their gameplay to the affective state of a player in order to deliver the optimal gaming experience. Indeed, an adaptive video game needs a continuous reading of the fun level to be able to respond to these changing fun levels in real time. While the predictive power of the presented classifier remains limited with a gain in the F1 score of 15% against random chance, it brings insight as to which physiological features might be the most informative for further analysis and discuss means by which low accuracy classification could still improve gaming experience.