The Effect of Feature Selection on Automatic Sleep Stage Classification Based On Multichannel EEG Signals

When diagnosing and treating sleep disorders, the manual classification of sleep stages is a time-consuming but crucial step, and the automation of this process has been a focus of recent research. Many kinds of research have been conducted on the automation of sleep stage classification. In this paper, we proposed the effect of feature selection on an automated system based on EEG signals, which was then followed by classification using various supervised classifiers such as Random Forest and SVM. The high dimensional FFT features were used to extract the characteristics of EEG for the classification of sleep stages. The EEG dataset is used from the Sleep-EDF dataset, which is freely available. The accuracy as the performance evaluator on the Random Forest model had gained the best value on 95.93%, 90.41%, 87.91%, 86.92%, and 84.86% and the SVM model reached 96.63%, 91.27%, 88.90%, 87.94%, and 87.94% for 2-6 state classification. Finally, in this proposed research the feature selection phase affects the model’s accuracy.