COMPARISON OF AUTOMATIC SLEEP STAGE SCORING METHODS USING LIMITED SCORING

Abstract

The diagnosis of various types of sleep disorders requires the experts to perform sleep stage scoring. However, it is an arduous and repetitive task and, therefore, an important candidate for automation. This work seeks to evaluate several scoring algorithms based on Machine Learning from the scientific literature. The comparison is performed with the same experimental design, using EEG, EOG and EMG signals from the polysomnographic records of the ISRUC-Sleep dataset. It is compared the precision, memory and speed of methods based on Linear Discriminant Analysis, Support Vector Machines, Random Forests and Neural Networks. As a result, several of the analyzed algorithms reach high levels of accuracy, exceeding 75%. Also, it is demonstrated that the accuracy can be raised to 85% by skipping the classification of doubtful epochs and still classify 65% of them.