Unsupervised Sleep Stages Classification Based on Physiological Signals

Abstract

Automatic sleep scoring has, recently, captured the attention of authors due to its importance in sleep abnormalities detection and treatments. The majority of the proposed works are based on supervised learning and considered mostly a single physiological signal as input. To avoid the exhausting pre-labeling task and to enhance the precision of the sleep staging process, we propose an unsupervised classification model for sleep stages identification based on a flexible architecture to handle different physiological signals. The efficiency of our approach was investigated using real data. Promising results were reached according to a comparative study carried out with the often used classification models.