An analysis of electrocardiograms in developing statistical learning models for instantaneous sleep quality and sleep potential prediction
MetadataShow full item record
The existence of a relationship between the electrical activity of the heart and the intricacies of sleep has been established but usually constrained to the context of diagnosis and more recently monitoring. However, in a more general sense, achieving a good night rest is a more present need. This study proposes a systematic framework that forecasts sleep quality in a circadian cycle by analyzing a continuous stream of Electrocardiograms (ECG). The time varying ECG signals are analyzed in the frequency domain and statistical learning models that forecast Sleep Quality (SQ) and outputs instantaneous Sleep Potential (SP) are developed. The attribute “Time to Sleep” (TTS) is introduced as a simple interpretation of a Sleep Potential (SP) value for easy clinical and mainstream adoption. The methodology aspires to describe a general framework which can be adapted to specialized cases and cater to specific biases by providing the appropriate training set to a supervised statistical learning model. This research explores the use of Support Vector Machines, Neural Networks and Multivariate Regression models for predicting Sleep Quality (SQ) as well as the instantaneous determination of Sleep Potential (SP). The Sleep Quality (SQ) is evaluated as the amount of time or data points in Slow Wave Sleep (SWS) and Rapid Eye Movement (REM) compared with the overall length of the sleep episode. Sleep Potential (SP) is evaluated as the temporal distance between the mean vector of sleep stages 1 - 5 in the training dataset and an input vector in stage 0 (Awake). The input vector encapsulates the Power Spectral Density (PSD) and frequency information.