Exploring the mechanisms of social jet lag in risk decision-making with machine learning models.

Risk decision-making describes how individuals evaluate uncertain options and accept potential losses in pursuit of rewards. This study used machine learning models to examine whether social jet lag and basic demographic or temporal variables can explain individual differences in risk decision-making. Using 407 valid responses from participants aged 17-23 years, we compared Linear Regression, Random Forest, and XGBoost regression models after grid-search tuning for the ensemble models and 5-fold cross-validation. The models showed limited predictive capability on the held-out test set (Linear Regression: R2 = 0.051, RMSE = 7.31; Random Forest: R2 = -0.004, RMSE = 7.52; XGBoost: R2 = 0.007, RMSE = 7.48). Linear Regression performed slightly better than the two nonlinear ensemble models, indicating that the available variables did not provide meaningful nonlinear predictive gains. Rather than presenting a deployable prediction tool, the study identifies the boundary conditions of using a small set of circadian and demographic features to model risk decision-making, and it highlights the need for richer behavioral, affective, and contextual data in future computational behavioral research.