EEG-Based Multi-Class Sleep Disorder Classification Using Deep Learning, Machine Learning and Neural Tangent Kernel

This paper presents a comprehensive approach for detecting insomnia using EEG signal analysis by combining traditional machine learning, deep learning, and Neural Tangent Kernel (NTK) methods. We analyze EEG data from three distinct classes: healthy individuals, insomnia patients, and subjects with sleep-disordered breathing (SDB). Our methodology involves sophisticated feature extraction techniques including Katz and Higuchi fractal dimensions, spectral analysis, and statistical measures. We evaluate several machine learning models, with Random Forest achieving $67.50 \%$ accuracy and XGBoost reaching 67.20% accuracy. A convolutional neural network (CNN) architecture is implemented for deep learning-based classification, achieving 72% accuracy. Additionally, we explore the NTK approach with kernel ridge regression, obtaining $68.34 \%$ accuracy. The comparative analysis demonstrates that ensemble methods and deep learning approaches provide robust performance for multi-class EEG classification. Our results suggest that EEG-based machine learning systems can serve as effective tools for assisting in insomnia diagnosis, potentially reducing reliance on subjective patient reports and lengthy clinical evaluations.