Drowsiness Detection using Instantaneous Frequency based Rhythms Separation for EEG Signals

Abstract

Drowsiness is the major cause of road accidents because it reduces the conscious level of the drowsy driver. The road accidents can be avoided by automatic detection of drowsiness state. In this paper, the electroencephalogram (EEG) rhythms-based features are proposed for the identification of drowsiness state. The Hilbert Huang transform computed instantaneous frequency is used for separation of rhythms from the empirical mode decomposition (EMD) provided intrinsic mode functions (IMFs). The separated EEG rhythms are used for the computation of time domain features namely mean, average amplitude change, coefficient of variation, trimean, activity, complexity, and neg-entropy. These features are tested on the variants of ensemble classifier for the classification of drowsiness and alertness states. In ensemble classifier variants, the bagged tree ensemble classification model provides best classification results as compared to other same dataset methods.