Related Experiment Video
Updated: Jul 4, 2026
Combining Computer Game-Based Behavioural Experiments With High-Density EEG and Infrared Gaze Tracking
Published on: December 16, 2010
Short-Horizon Prediction of Player Well-Being from Gameplay Behavior and Psychological Signals
Eleftherios Vouzis1, Nikolaos Vouzis1, Paschalina Lialiou1,2
1Department of Digital Systems, University of Piraeus.
None:
Understanding how video gaming relates to psychological well-being remains challenging, as prior studies rely on cross-sectional designs prone to construct overlap and limited longitudinal sampling. This study develops a player-agnostic machine learning framework to predict short-horizon well-being from preceding gameplay behavior.
Methods:
Using the PowerWash Simulator Open Dataset (8,372 players), 34 features are extracted from 2-hour gameplay windows preceding each response. A tertile-based binary classification (low vs high well-being) is constructed, excluding the middle group, with thresholds computed within each training fold to prevent leakage. Five-fold GroupKFold validation ensures zero player overlaps. SHAP is used for interpretability.
Results:
LightGBM achieved 81.1% accuracy (AUC=89.9%) without prior well-being features. SHAP identified enjoyment, focus, and autonomy as dominant predictors.
Conclusion:
Short-term gameplay patterns demonstrate strong predictive power for well-being, enabling interpretable, real-time monitoring systems grounded in behavioral and psychological indicators.
