JMIR Publications

ABSTRACT

Background:

Depression, marked by persistent sadness and loss of interest in daily activities, greatly reduces quality of life. Early detection is vital for effec- tive treatment and intervention. While many studies use wearable devices to classify depression based on physical activity, these often rely on intrusive meth- ods. Additionally, most depression classification studies involve large participant groups and use single-stage classifiers without explainability.

Objective:

In this study, we aim to assess the feasibility of classifying depression using non-intrusive WiFi-based motion sensor data, employing a novel machine learning model on a limited number of participants. We also conduct an explain- ability analysis to interpret the model’s predictions and identify key features associated with depression classification.

Methods:

We recruited adults aged 65+ through online and in-person meth- ods, supported by a McGill University healthcare facility directory. Participants provided consent, and we collected six months of activity and sleep data via non-intrusive WiFi-based sensors, along with Edmonton Frailty Scale (EFS) and Geriatric Depression Scale (GDS) data. For depression classification, we pro- posed a three-stage ML model with feature selection, dimensionality reduction, and classification, evaluating various model combinations using accuracy, sensi- tivity, precision, and F-score. SHAP and LIME were used to explain the model’s predictions.

Results:

A total of six participants were enrolled in this study; however, two withdrew later due to internet connectivity issues. Among the four remaining par- ticipants, three were classified as not having depression, while one was identified as having depression. The most accurate classification model, which combined Sequential Forward Selection (SFS) for feature selection, Principal Component Analysis (PCA) for dimensionality reduction, and a Decision Tree (DT) for classification, achieved an accuracy of 87.5%, sensitivity of 90.0%, and preci- sion of 88.3%, effectively distinguishing individuals with and without depression. The explainability analysis revealed that the most influential features in depres- sion classification, in order of importance, were ”average sleep duration”, ”total number of sleep interruptions”, ”percentage of nights with sleep interruptions”, ”average duration of sleep interruptions”, and ”Edmonton Frail Scale (EFS)”.

Conclusions:

The findings from this preliminary study demonstrate the feasibil- ity of using WiFi-based motion sensors for depression classification and highlight the effectiveness of our proposed three-stage machine learning model, even with a small sample size. These results suggest the potential for further research with a larger cohort for more comprehensive validation. Additionally, the non-intrusive data collection method and model architecture proposed in this study offer promising applications in remote health monitoring, particularly for older adults who may face challenges in using wearable devices. Furthermore, the importance of sleep patterns identified in our explainability analysis aligns with findings from previous research, emphasizing the need for more in-depth studies on the role of sleep in mental health, as suggested in the XML study.