Nighttime Sleep Duration Prediction for Inpatient Rehabilitation Using Similar Actigraphy Sequences

Wearable Device-Independent Next Day Activity and Next Night Sleep Prediction for Rehabilitation Populations

“…We demonstrated such prediction models trained with data collected from 44 inpatient rehabilitation subjects can achieve NRMSE values near 9% for daytime physical activity prediction and near 11% for nighttime sleep duration prediction. These results were an expansion over own prior work with data from a single sensor device [36] . For future work, we plan to continue growing our sample size to provide additional historical data sequences for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$K$ \end{document} NN to select from.…”

“…We demonstrated such prediction models trained with data collected from 44 inpatient rehabilitation subjects can achieve NRMSE values near 9% for daytime physical activity prediction and near 11% for nighttime sleep duration prediction. These results were an expansion over own prior work with data from a single sensor device [36]. For future work, we plan to continue growing our sample size to provide additional historical data sequences for KNN to select from.…”

“…To predict TAM, TIM, and TST based on the extracted features, we utilized a 100-tree random forest, employing a mean squared error feature-selection criterion, enhanced with bagging, and without a maximum depth value [38] . We chose a random forest regression algorithm because it is an ensemble approach that exhibited low variance and yielded high prediction accuracy in our previous work predicting nighttime sleep duration [36] . Furthermore, the random forest regressor outperformed other regression models we explored, including \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$K$ \end{document} -nearest neighbors regressors, support vector regressors, and neural networks.…”

“…Machine learning models trained with the collected daytime physical activity data were used to classify good and poor sleep efficiency with an area under the receiver operating curve of 0.9449. In our recent work, we expanded this research to investigate the applicability of sleep prediction for individuals with sleep disorders [36]. For this study, we deployed AMI MotionLogger devices in an inpatient rehabilitation setting.…”

“…For the AMI dataset, we analyzed data from 17 subjects (age 64.11 ± 17.05 years; 11 females and 6 males), for which the data collection periods ranged from 9 days to 30 days [36]. For the Philips dataset, we analyzed data from 22 subjects (age 63.96 ± 17.93 years; 5 females and 17 males), for which the data collection periods ranged from 4 days to 30 days.…”