Fig. 2. Stressful virtual reality scenario and model prediction of cardiac stress responses from behavior.

a Scenario 3: Dark maze, aimed at eliciting heightened physiological responses. b Heart rate (HR) and heart rate variability (HRV, the RMSSD formula) over time (three blocks of 30 s each) during dark maze exploration. N = 135 participants examined over three consecutive blocks of 30 s (S1, S2, and S3). Data are presented as mean values ± SEM (vertical lines). A repeated measures analysis of variance (rm-ANOVA) was performed for both HR and HRV data sets. Post hoc tests were performed with two-sided paired t tests, with p values corrected for multiple comparisons (three comparisons) with the Holm correction. ***p value <0.001. Exact p values and statistics are presented in Supplementary Information section “Statistics for physiology in exploration scenarios”, Tables S6–7. Correlation between the XGBoost model’s prediction and the integrated HRV index (iHRV) for the test set c and train + test set d. e SHAP values for the trained model, calculated for each subject. The nine most discriminating features (from the 18 used by the model) were the minimum distance to the corners of the empty room [Corner dist (er)], vertical acceleration in the elevated alley [Vert acc (ea)], ratio between time in the center and time in the periphery of the empty room [Ratio time (er)], time spent on the narrowest ledge of the elevated alley [Time narrow (ea)], movement focus in the empty room [Focus (er)], variability in the stride speed in the empty room [Stride speed var (er)], number of head scans while walking in the empty room [Walking head scan (er)], time spent on the starting board in the elevated alley [Time board (ea)], maximal longitudinal distance reached in the elevated alley [Distance long (ea)].