Detection of Excessive Daytime Sleepiness in Resting-State EEG Recordings: A Novel Machine Learning Approach Using Specific EEG Sub-Bands and Channels

Abstract

Excessive Daytime Sleepiness (EDS) is a threat to the safety of the patient as well as others and can lead to higher rates of mortality. While current tests for EDS are time-consuming, expensive, and inaccurate, modern data-driven approaches can significantly improve diagnosis and treatment quality. Using an improved machine learning (ML) method in combination with the unfolding of electroencephalogram (EEG) bandwidths in a fine-graded equidistant 99-point spectrum, we successfully developed and evaluated a fast classification model to correctly diagnose EDS. Our model achieved a good level of accuracy of 87.00 percent on completely unseen resting-state EEG data. Our model is one of the first to detect long term EDS instead of short-term drowsiness from EEG data and significantly extends related work on the detection of drowsiness by considering the local occurrence of differences in spectral power.