JMIR Publications

ABSTRACT

Background:

Affective states are important aspects of healthy functioning; as such, monitoring and understanding affect is an important aspect in the assessment and treatment of mood-based disorders. Recent advancements of wearable technologies have increased the utilization of such tools in detecting and accurately estimating mental states (e.g., affect, mood, stress, etc.), offering comprehensive and continuous monitoring of individuals over time.

Objective:

Previous attempts to model an individual’s mental state relied on subjective measurements or the inclusion of only a few objective monitoring modalities (e.g., smartphones). Our aim in the current study is to investigate the capacity of monitoring affect using fully objective measurements. We conduct a comparitively long term (12 month) study with a holistic sampling of participants' mood, including 20 affective states.

Methods:

Longitudinal physiological data (such as sleep and heart rate) as well as daily assessments of affect were collected using three modalities (i.e. smart phone, watch, and ring) from 20 college students over the course of a year. We examined the difference between distributions of data collected from each modality along with differences between their rates of missingness. Seven out of the twenty participants provided us with 200 or more days worth of data, and we used this for predictive modeling setup. Distributions of positive and negative affect (PA and NA) of the selected 7 participants were observed. For predictive modeling, we assessed the performance of different machine learning models, including random forests (RF), support vector machine (SVM), multilayer perceptron (MLP), and K-nearest neighbor (KNN). We also investigated the capability of each modality in predicting mood and the most important features of PA and NA RF models.

Results:

RF was the best performing model in our analysis and performed mood and stress (nervousness) prediction with ~81% and ~72% accuracy, respectively. PA models resulted in better performance compared to NA. The order of most important modalities in predicting PA and NA was smart ring, phone and watch respectively. SHAP analysis showed that sleep and activity related features were the most impactful in predicting PA and NA.

Conclusions:

Generic machine-learning-based affect prediction models, trained with population data, outperform existing methods, which use the individual’s historical information. Our findings indicated that our mood prediction method outperformed the existing methods. Additionally, we found that sleep and activity level were the most important features for predicting next-day PA and NA, respectively.