Machine learning models can tell us a lot about how patients sleep, according to new research published in PLOS One. And it’s much less obtrusive than prior methods.
“People spend almost one-third of their lifetime sleeping,” wrote Xingyun Liu, Chinese Academy of Sciences in Beijing, and colleagues. “Adequate sleep is an important prerequisite for good health, while bad sleep can result in bad moods, inattention, fatigue, cardiovascular disease and even mortality. Currently, people pay considerable attention to their sleep quality.”
The authors noted that self-reported questionnaires and sleep-tracking smartphones applications have been used in recent years to monitor sleep patterns—but both methods have their own disadvantages. Questionnaires can be unreliable, for instance, and smartphones can be “inconvenient” and “obtrusive.”
Liu et al. turned to a combination of machine learning models and Microsoft Kinect to capture data about patients’ gait—how they walk—and evaluate the quality of their sleep. Fifty-six first-year postgraduate students from a single university were recruited as participants, a group that included 24 men and 32 women. Participants were then asked to move freely for two minutes as two Kinect sensors collected gait data. Data collection was different than previous studies, the team explained, because “gait patterns” were observed instead of “gait speed.”
The participant data was then used with a set of machine algorithms that included linear regression (LR), simple linear regression (SLR), Gaussian processes (GP), epsilon-support vector regression (E-SVR) and nu-support vector regression (N-SVR). Models were trained using 10-fold cross validation.
“To be specific, we randomly selected nine-tenths of the dataset as the training data and the remaining data as the testing data, and then we repeated the process ten times for every model,” the authors wrote. “This method can avoid problems such as overfitting or selection bias to some degree.”
Overall, GP achieved the best score with a correlation of 0.78.
“The correlation coefficient value of 0.78 indicates a high positive correlation, which is relatively rare according to previous studies,” the authors explained. “This result is consistent with an earlier study showing that sleep is associated with gait and demonstrates that gait patterns can reveal sleep quality quite well. More importantly, a real-time prediction of human health status (sleep quality) scores can be implemented by using Kinect.”
Michael has more than 16 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.