Classification of Sleep Stages using Neural Networks

Resumo

Sleep stages are considered important indicators for diagnosing neurological and psychiatric illnesses,
so sleep disturbances have a major impact on a person's well-being. However, many sleep status analyzes are done
visually by professionals in the field, which can be a slow task. For this reason, studies on sleep states automatically
are of great importance in diagnosing neurological disorders quickly. In this article, an approach using the variance
of the electroencephalogram (EEG) as an input to a multilayered neural network was used to classify sleep states.
The EEGs were obtained from the Sleep-EDF Database Expanded database, obtained from two healthy patients,
the first containing 7,950,100 records (22 total hours), and the second 8,490,100 (23 total hours), being extracted,
filtered and subsequently classified automatically into five sleep states. A neural network used was fully connected

with two intermediate layers and 12 neurons in each layer, with a learning rate of 0.01. The accuracy of the five-
stage classification was 94.3%. These results showed that the proposed algorithm is favorable to obtain a

classification of the stages with good precision in comparison to the other results found in the literature.