Abstract:
Existing automated sleep stage classification techniques tend to focus on NREM and REM
but merrily ignore the Wake phase which is essential for studying insomnia. Our aim is to
bring a simple and comprehensible machine learning method to fill this gap of the precise
expressions of the Wake stages. Our Gradient Boosting algorithm, compared to other
algorithms, has shown 91.08% accuracy in classifying NREM, REM, and the Wake stages
from single channel EEG signals. In the meantime, XGBoost demonstrated fantastic
performance delivering 90.99% accuracy. To address Wake data scarcity, we integrated
SMOTE to enhance overall classifier effectiveness. AdaBoost achieved 81.29% accuracy,
but Gradient Boosting did better as it outperformed the baseline by having 89.47% on
unseen data. The results presented here help building a clinically intuitive, very accurate
instrument for personal sleep monitoring, which is directly applicable to home healthcare
and insomnia therapy.
