Passive Sensing for Mental Health Monitoring: A Scoping Review of Machine Learning with Wearables and Smartphones (Preprint)
Authors:
- ShiYing ShenSSShiYing Shen
- Wenhao QiWQWenhao Qi
- Jianwen ZengJZJianwen Zeng
- Sixie LiSLSixie Li
- Xin LiuXLXin Liu
- Xiaohong ZhuXZXiaohong Zhu
- Chaoqun DongCDChaoqun Dong
- Bin WangBWBin Wang
- Yankai ShiYSYankai Shi
- Jiani YaoJYJiani Yao
- Bingsheng WangBWBingsheng Wang
- Xiajing LouXLXiajing Lou
Preprint
- 7 May 2025
- preprint
- Published by JMIR Publications Inc. in JMIR Preprints
Abstract
BACKGROUND Mental health issues have become a significant global public health challenge. Traditional assessments rely on subjective methods with limited ecological validity. Passive sensing via wearable devices and smartphones, combined with machine learning (ML), enables objective, continuous, and noninvasive mental health monitoring. OBJECTIVE This study aims to provide a comprehensive review of the current state of passive sensing-based and machine learning (ML) technologies for mental health monitoring. We summarize the technical approaches, reveal the association patterns between behavioral features and mental disorders, and explore potential directions for future advancements. METHODS Following the PRISMA-ScR guidelines, we searched seven major databases (Web of Science, PubMed, IEEE Xplore, etc.) for studies published between 2015 and 2025. A total of 42 studies were included. Information was extracted from dimensions such as data collection, preprocessing, feature engineering, ML methods, and validation, integrating the association patterns between behavioral features (e.g., sleep, activity, social interaction) and mental disorders (e.g., depression, anxiety). RESULTS The study found that the most commonly used digital biomarkers were heart rate (n=28), movement index (n=25), and step count (n=17), which were significantly associated with mental disorders such as depression and anxiety. Deep learning models (e.g., CNN, LSTM) performed exceptionally well in processing time-series data. However, traditional methods (e.g., random forest, XGBoost), due to their higher interpretability, remain widely adopted. Current studies face challenges such as small sample sizes (median = 60.5 participants), short data collection periods (45.24% of studies had data collection periods of less than 7 days), and limited device variety (76.19%). Additionally, only one study conducted external validation, limiting the clinical generalizability of the models. On the ethical front, only a few studies (14.29%) explicitly mentioned data anonymization, highlighting the need for enhanced privacy protection and algorithm fairness. CONCLUSIONS The combination of passive sensing technologies and machine learning offers innovative solutions for mental health monitoring. However, key challenges, including data quality, model generalization, and ethical standards, need to be addressed before clinical translation. Future research should focus on large-scale longitudinal data collection, multimodal data integration, algorithm optimization, and interdisciplinary collaboration to drive the widespread adoption and clinical application of these technologies.Keywords
This publication has 92 references indexed in Scilit:
- Hopping-mean: an augmentation method for motor activity data towards real-time depression diagnosis using machine learningMultimedia Tools and Applications, 2024
- Advancing ADHD diagnosis: using machine learning for unveiling ADHD patterns through dimensionality reduction on IoMT actigraphy signalsInternational Journal of Information Technology, 2024
- Utilizing deep convolutional neural architecture with attention mechanism for objective diagnosis of schizophrenia using wearable IoMT devicesMultimedia Tools and Applications, 2023
- Depress-DCNF: A deep convolutional neuro-fuzzy model for detection of depression episodes using IoMTApplied Soft Computing, 2022
- A study on multi-class anxiety detection using wearable EEG headbandJournal of Ambient Intelligence and Humanized Computing, 2021
- An empirical comparison and evaluation of minority oversampling techniques on a large number of imbalanced datasetsApplied Soft Computing, 2019
- Causability and explainability of artificial intelligence in medicineWIREs Data Mining and Knowledge Discovery, 2019
- Digital PhenotypingJAMA, 2017
- Social neuroscience in psychiatry: pathways to discovering neurobiological risk and resilienceWorld Psychiatry, 2014
- Basic and complex emotions in depression and anxietyClinical Psychology & Psychotherapy, 2007