An accurate sleep staging system with novel feature generation and auto-mapping

Traditional sleep monitoring conducted in professional sleep labs and scored by sleep specialist is costly and labor intensive. Recent development of light-weight headband EEG provides possible solution for home-based sleep monitoring. This study proposed a machine learning approach for automatic sleep stage detection. A set of effective and efficient features are extracted from EEG data. The utilization of a collection of well annotated sleep data ensures the quality of learning model. A feature mapping algorithm is proposed to map the feature spaces generated from EEG data acquired through different electrodes. We collected headband EEG data for 1 hour naps in experiments conducted in our sleep lab. Preliminary result shows that sleep stages detected by proposed method are highly agreeable with the sleepiness score we obtained.