کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Comparative Analysis of Classical Sleep Staging Classifiers on In-ear EEG Dataset

نویسندگان :

Ali Yazdanifar¹ Fatemeh Elahi² Shervin Mehrtash³ Alireza Akhavanpour⁴ Reza Ebrahimpour⁵

1- دانشگاه تربیت دبیر شهید رجایی 2- پژوهشگاه دانش‌های بنیادی 3- دانشگاه صنعتی شریف 4- دانشگاه تربیت دبیر شهید رجایی 5- دانشگاه صنعتی شریف

کلمات کلیدی :

sleep-staging،wearable electrode،Light-weight classifier،In-ear EEG

چکیده :

sleep disorders affect millions of individuals worldwide. They are linked to metabolic, cardiovascular, and neurocognitive complications and can disrupt many aspects of daily life. Disturbances in sleep stages often underlie these conditions. For this reason, accurate monitoring of sleep stages is essential for understanding and diagnosing sleep disorders. Polysomnography remains the gold standard for sleep assessment. However, its dependence on laboratory settings and specialized equipment limits its suitability for routine or long-term monitoring. In-ear EEG is a practical alternative for wearable sleep monitoring, enabling stable, gel-free, and self-administered recordings with minimal interference in normal activities. Sleep recordings are still commonly assessed through manual inspection or simple statistical methods. More recent studies have moved toward automated approaches, offering improved diagnostic performance. In this study, we evaluated the performance of traditional light-weight classifiers in a five-class sleep staging task based on AASM sleep scoring guidelines. In-ear EEG recordings from six healthy subjects were analyzed across full-night sleep sessions. A wide range of temporal, spectral, and nonlinear features was extracted and categorized into 20 feature groups. Classification performance was examined using both within-subject and cross-subject validation schemes. The k-nearest neighbors classifier achieved the highest accuracy under within-subject and cross-subject validation regimes, with 81.7% accuracy averaged among never-seen-before test subjects. These results highlight the feasibility of sleep stage classification using simple models and in-ear EEG, supporting their use in practical and wearable sleep monitoring applications.