Work

Development of personalized machine learning models for predicting panic attacks in college students

Year: 2026

Type: article

Abstract: Panic attacks (PAs) are unexpected episodes of intense fear and anxiety, accompanied by cognitive and physical symptoms that significantly impair day-to-day functioning. This study proposes a machine learning approach to predict PA risk in individuals across mental health conditions, enabling timely and potentially preventive mental health intervention. We demonstrate the efficacy of the approach in a longitudinal sample of behavioral and physiological data collected using Oura rings from 42 college students over six months. Of personalized models, 20 achieved performance significantly better than random chance with a mean PR AUC of 0.85, outperforming those based on self-reported distress scores (mean PR AUC = 0.79 for n=21) and self-reported daily stressor counts (mean PR AUC = 0.94 for only n=7), which both require active user engagement. Low Jaccard similarity (12.6%) of key features for predicting PAs across personalized Oura models, and lower performance of the generalized cohort model (PR AUC = 0.40), further supports the need for personalized modeling approaches. Post-hoc analyses of personalized models suggest that only two PAs are needed to enable suitable model performance. Notably, for the 22 personalized Oura models with poor performance, better than random chance models were identified for nearly all (n=20) through an innovative cross-participant approach. Collectively, findings suggest that passively collected daily measurements of sleep, activity, and physiology can be used to predict PAs and could be leveraged to identify individually tailored lifestyle interventions for reducing PA risk.

Source: Frontiers in Digital Health

DOI: 10.3389/fdgth.2026.1978650