Predicting Adolescent Depressive Relapse via LSTM Modeling of Smartphone Telemetry
Journal of Adolescent and Youth Psychological Studies · 2026 · 7 (3) · 1-10
DOI: 10.61838/kman.jayps.3142 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Adolescent and Youth Psychological Studies, 2026. Verfasst von Elena Stoyanova und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Objective
To evaluate the predictive efficacy of a Long Short-Term Memory (LSTM) neural network in forecasting the likelihood of an impending major depressive relapse among a remitted adolescent cohort using longitudinal, passively collected smartphone behavioral telemetry. Methods and Materials: This longitudinal observational study tracked a cohort of N=427remitted adolescents in Morocco over an eighteen-month period. Behavioral data was continuously and unobtrusively collected using a custom passive sensing smartphone application that recorded high-resolution telemetry, including geospatial mobility (GPS), accelerometer-based physical activity, total screen time, and keystroke dynamics. Clinical mental health status was evaluated bi-weekly via Ecological Momentary Assessment (EMA). An LSTM recurrent neural network, designed to capture temporal dependencies in time-series data, was trained on these sequential behavioral inputs to predict the future probability of a clinical depressive relapse.
Findings
During the observation window, N=142adolescents experienced a formal depressive relapse. The LSTM predictive model achieved strong performance metrics, forecasting an impending relapse with an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.88over a 14-day predictive horizon. The model demonstrated a sensitivity of 0.83 and a specificity of 0.81. A subsequent feature ablation analysis revealed that severe reductions in geospatial mobility variance and profound disruptions in circadian sleep patterns—quantified by acute spikes in nocturnal device usage—served as the most significant digital biomarkers driving the algorithm’s predictive accuracy.
Conclusion
Passively collected smartphone telemetry contains highly predictive, sequential signals of clinical deterioration that precede self-reported depressive symptoms. Utilizing advanced LSTM architectures to model these digital footprints provides a highly accurate, objective mechanism for forecasting adolescent depressive relapses, offering a critical window for proactive, digitally-augmented preventative psychiatric interventions.
Zitieren
APA
Stoyanova, E., Amrani, R. E., & Hmeyada, R. (2026). Predicting Adolescent Depressive Relapse via LSTM Modeling of Smartphone Telemetry. Journal of Adolescent and Youth Psychological Studies, 7(3), 1-10. https://doi.org/10.61838/kman.jayps.3142
Vancouver
Stoyanova E, Amrani RE, Hmeyada R. Predicting Adolescent Depressive Relapse via LSTM Modeling of Smartphone Telemetry. Journal of Adolescent and Youth Psychological Studies. 2026;7(3):1-10. doi:10.61838/kman.jayps.3142
BibTeX
@article{Stoyanova2026,
author = {Stoyanova, Elena and Amrani, Rachid El and Hmeyada, Rezki},
title = {Predicting Adolescent Depressive Relapse via LSTM Modeling of Smartphone Telemetry},
journal = {Journal of Adolescent and Youth Psychological Studies},
year = {2026},
volume = {7},
number = {3},
pages = {1--10},
doi = {10.61838/kman.jayps.3142}
}
EUVIMED