Depression is a significant mental health condition that results in persistent sadness together with diminished interest and inability to experience joy. The condition has become a major concern which particularly affects elderly people. Early intervention and care offer essential benefits to mental health outcomes while decreasing health service requirements for both patients and medical systems.
The standard depression assessment protocols require professional consultations and knowledge training for healthcare providers while requiring them to conduct questionnaire interviews that are time-consuming and costly. Implementing these assessment techniques causes additional strain on elderly patients due to their required frequent visits to health facilities and limits the accessibility of clinical data collection. Together with their preference for home residence, many older adults find remote monitoring highly appealing.
The relationship between physical activity and mobility directly affects the understanding of depression because these two elements are directly linked. Advancements in the Internet of Things (IoT) technology enable continuous, real-time physical activity monitoring through remote tracking software. Researchers conduct studies where wearable technology monitors the connection between physical activity and depression. The integration of artificial intelligence enables machine learning algorithms to become better depression prediction tools by processing physical activity information.
Participants for this study were recruited using an online and home-based enrollment method, which received support from a McGill University health facility directory. A deeper understanding was obtained from adults aged 65 and older by collecting their consent approval, followed by a six-month tracking period. This study involved Wi-Fi sensors, the Edmonton Frailty Scale, and Geriatric Depression Scale measurements.
The researchers built the Home-Based Older Adults’ Depression Prediction (HOPE) model using feature selection and classification with dimensionality reduction steps. Different model configurations were explored by testing their performance metrics, including accuracy, sensitivity, precision, and F1-score along the Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanation techniques. These techniques provided insights into how the model reached its predictions.
The research team used digital communication and in-person interactions to find participants for the study. The study information distribution included emails together with social media and electronic posters. After interested participants provided consent, the research staff installed the monitoring equipment with support. The participants received rewards through an e-gift card. Some participants were excluded from the study based on two key conditions: (1) physical or mental impairments that affect balance or mental state and the use of mobility aids such as canes and (2) active substance use, as it could influence physical activity levels. The study included participants who had previously engaged in substance use but had stopped using them.
Internet connectivity issues caused two participants to drop out of the study from a total participant count of six. Of the four who remained, three were not diagnosed with depression, while one participant was. The depression classification model that included feature selection, dimensionality reduction, and decision tree classification methods achieved 87.5% accuracy with 90% sensitivity and 88.3% precision. This modeling system effectively distinguished between individuals with and without depression. The explainability analysis identified “average sleep duration” and “total number of sleep interruptions” as well as “proportion of nights with sleep interruptions,” “average sleep interruption duration,” and “Edmonton Frailty Scale” to be the most significant factors in depression recognition.
This result of the small-scale study demonstrates that Wi-Fi sensor technology can successfully identify symptoms of depression in people. The HOPE machine learning model results delivered satisfactory predictions with minimal participation numbers. Research into this subject using increased participant numbers will likely yield better accuracy in the results. The modest data collection approach and model characteristics provide strong potential for distant health tracking services that can reach older adults who prefer non-wearable monitoring methods. Additionally, research demonstrates that sleep patterns play a critical role in model predictions, reinforcing the need for further studies on the connection between sleep and mental health.
References: Nejadshamsi S, Karami V, Ghourchian N et al. Development and feasibility study of HOPE model for predicting depression among older adults using Wi-Fi-based motion sensor data: machine learning study. JMIR Aging. 2025;8:e67715. doi:10.2196/67715
