Impact of Stress on Sleep Levels: A Comparative Machine Learning Study Based on Wearable Data

Inadequate sleep is a significant factor in chronic conditions such as cardiovascular disease, diabetes, and mental health disorders, and includes not just sleep duration but also the quality of sleep. Although addressing sleep disorders is complex, mitigating their physical and psychological consequences through effective monitoring and control measures is achievable using wearable devices. While there have been a variety of successful works using wearable devices to classify sleep levels, this study investigates the feasibility of machine learning (ML) approaches based on wearable data to predict different levels of sleep by considering physiological parameters including stress. Stress has long been established as one of the most crucial physiological parameters influencing sleep levels. In this paper, we leveraged data collected from wrist-worn devices to classify the levels of sleep using two different datasets: one considering the level of stress and the other excluding it. The objective is to achieve optimal classification performance by employing a diverse set of seven ML-based classifiers, including Support Vector Machines (SVM), Logistic Regression, K-Nearest Neighbors (KNN), Decision Trees, Random Forests, Naive Bayes, and XGBoost. By comparing the performances of the models on these two datasets, we observed that the model incorporating stress inputs exhibited superior accuracy in predicting sleep, highlighting the significant role of stress in sleep levels. Moreover, the findings from the models and their comparison indicate that the Random Forest and XGBoost algorithms outperform other ML methods, exhibiting an impressive accuracy of 95 % in accurately predicting different sleep levels.