NJIT
Passive interpretable deep learning from cell phone activity for optimal patient care
Abstract
dc:description.abstract<p>The majority of individuals in need of mental healthcare fail to seek it, and the diagnosis of those who do is driven predominantly by the patient's subjective self-assessment. There is a need for mental illnesses trackers that measure psychological and physiological symptoms and provide alerts and interpretable data for therapeutic interventions. We present a system that complements existing protocols to detect distress in mental health, passively monitor patients, and quickly alert authorized parties on possible mental health distress. We evaluate modeling a person's mental health from the sensors and activity logs of her/his cell phone using logistic regressions, LSTM, and LSTM with attention. The system distinguishes healthy from clinically depressed patients with 81% accuracy and provides explanations to guide practitioners. The system works without the need for social interactions and active labeling. Moreover, the system paves the way toward interpretable mental health prediction models.</p>
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy in Business Data Science - (Ph.D.)
- Discipline thesis:degree_discipline
- School of Management
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Ayranci, Pelin
- 0000-0002-6094-5095
- Contributors dc:contributor
-
- Cesar Bandera
- Hai Nhat Phan
- Michael A. Ehrlich
Subjects
dc:subject × 7Identifiers
dc:identifier.*- Repository record dc:identifier
- https://digitalcommons.njit.edu/dissertations/1850