Automated Insomnia Detection Using Multi-Channel EEG Signals — A Comparative Study

Insomnia is a prevalent sleep disorder that adversely impacts cognitive, emotional, and physical health. Traditional diagnostic methods, such as polysomnography and self-reported sleep assessments, are constrained by high cost, invasiveness, and subjectivity, highlighting the need for automated, scalable solutions. This study presents a comparative framework for insomnia detection using multi-channel EEG signals, integrating both handcrafted feature-based machine learning models and raw-segment deep learning architectures. EEG recordings were preprocessed through filtering, resampling, segmentation, and subject-wise normalization, followed by feature extraction via wavelet decomposition and entropy measures. Classification models, including Random Forest, Support Vector Machine, XGBoost, Multilayer Perceptron (MLP), and Convolutional Neural Network (CNN), were systematically evaluated using rigorous subject-disjoint cross-validation strategies. The MLP achieved the highest subject-level accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">9 4. 1 2 %</tex>, outperforming traditional classifiers and demonstrating the advantages of deep feature modeling. The results confirm the efficacy of deep learning in capturing complex EEG patterns associated with insomnia. This framework demonstrates strong potential as a non-invasive, accurate tool for early insomnia detection, providing a foundation for further advancements through expanded datasets, multimodal biosignals, and enhanced interpretability.