Exploring the psychophysiological predictors of performance under stress: insights from a machine learning approach

Abstract Psychological and physiological responses co-occur during stressful tasks and jointly influence performance. Yet, research on stress–performance links often focuses on single parameters in isolation, overlooking their interrelations. The current study addressed this gap using machine learning—specifically random forest regression—to identify key psychophysiological predictors of communication performance when considered simultaneously. Participants were 229 medical students who engaged in breaking bad news encounters with simulated patients. We assessed neuroendocrine and cardiovascular activity, mood states, emotion regulation strategies and stress appraisals as predictors and communication performance as the outcome. Results revealed suppression of unpleasant feelings as the strongest predictor, with greater suppression linked to poorer performance. Physiologically, better performance was associated with moderate decreases in heart rate variability, increases in cortisol and decreases in stroke volume. Analysis of joint effects indicated that suppression was especially detrimental under high physiological reactivity and among lower-performing individuals, suggesting cognitive overload under these conditions. The findings pinpoint suppression of unpleasant feelings as a maladaptive stress coping strategy, while physiological stress responses within the observed ranges appeared to enhance performance. Ultimately, addressing maladaptive emotion regulation and leveraging assessable physiological indicators—particularly heart rate variability reactivity—could inform training programmes to improve performance under stress.