Discrete Wavelet Transform Coefficients for Drowsiness Detection from EEG Signals

Abstract

This paper proposes an effective approach to detect drowsiness from EEG signals by using Discrete Wavelet Transform (DWT) coefficients as features. The majority of drowsiness detection systems extract features using FFT to calculate the power spectral density or the DWT to calculate entropy from EEG sub-bands. Although these techniques excel in capturing valuable features in the frequency domain, they omit temporal details essential to the analysis of EEG signals. These details are integrated into coefficients indicating the correlation between the wavelet function and the EEG signal at different times. In our work, we perform a time-frequency analysis of EEG signals using DWT coefficients to preserve this temporal context. Furthermore, the study explores the influence of time segment size on system performance. Subsequently, we determine the most suitable technique to minimize input feature redundancies. Our approach employs just two EEG electrodes, C3 and C4, mirroring common setups for detecting wakefulness and drowsiness. Four classifiers were assessed: decision tree, random forest, multilayer perceptron, and support vector machine. The findings reveal that DWT coefficients enhance drowsiness detection performance, surpassing previous methods.