Analysis of sleep stages with support vector machine

Abstract

Paramount to a healthy living style, and significant consequences in the absence of, sleep has been a topic of research for longer than a century. With the increasing number of sleep deprived individuals, sleep studies have become more crucial than ever.This thesis researches an approach to the validation of the self applied somnography, an emerging sleep study arrangement capable of providing flexible and more comfortable sleep studies. Using sleep recordings from two different somnograph arrangements, the supervised machine learning model - support vector machine - is trained and tested in the classification of NREM and REM. Results show that sleep disorders can significantly affect the accuracy of a support vector machine, as wake periods do not fit the labels of NREM and REM. Prominent results show that the SAS data set obtains higher accuracies than the PSG for every kernel used, with the polynomial kernel obtain the highest for both data sets.