Objective: We test whether data collected from daily passive smartphone sensing and self-reports can predict daily significant psychological distress with meaningful potential clinical utility, comparing across best-in-class machine learning algorithms. Methods: In this prospective cohort study, 169 community-dwelling adults completed baseline surveys followed by 7 days of passive smartphone sensing (accelerometer, GPS, power state) and end-of-day Brief Symptom Inventory-18 (BSI-18) assessments. Cases were defined as days (total n = 1,062) where participants’ psychological distress measured wi
Classical machine learning versus transformers for predicting daily psychological distress from passive smartphone sensing and self-report: a prospective cohort study
calendar_today
November 6, 2026
person
Griss Salas, I., Lam, W.-Y., Kilshaw, R. E., Baucom, B. R. W., de Lacy, N.
domain
medrxiv