EEG-based Sleep Deprivation Classification: A Performance Analysis of Channel Selection on Classifier Accuracy

Abstract

This study analyses the effect of electroencephalogram (EEG) channel selection on the classification accuracy of sleep deprivation using four distinct classifiers: Random Forest (RF), k-Nearest Neighbours (k-NN), Support Vector Machine (SVM), and Artificial Neural Network (ANN). In this study, the EEG data from ten male individuals in good health were collected. Two distinct sets of EEG channels—a limited frontal channel set (Fp1, Fp2) and a thorough 19-channel set—were used to compare the performance of the classifiers. According to our findings, the k-NN classifier produced the greatest classification accuracy of 99.7% when applied to the 19-channel EEG signals. In contrast, both SVM and ANN classifiers were able to obtain the greatest accuracy of 94% with the frontal channels. Though there are not many gaps, these results imply that employing a larger range of EEG channels greatly improves the classification accuracy of sleep deprivation. The present study emphasizes the significance of channel selection in EEG-based sleep deprivation investigations by showcasing the significant advantages of full EEG signal capture over minimum channel configurations.