Drowsiness Analysis Using Common Spatial Pattern and Extreme Learning Machine Based on Electroencephalogram Signal

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Driver drowsiness is a significant cause of accidents.
• Conventional drowsiness detection systems have limitations.
• Wearable electroencephalogram (EEG) offers a lightweight and user-friendly alternative.

Purpose of the Study:

• To develop an effective drowsiness detection system using wearable EEG.
• To improve the accuracy and efficiency of sleepiness detection in drivers.

Main Methods:

• EEG signals acquired using EMOTIV Epoc+.
• Decomposition of EEG into frequency bands (delta, theta, alpha, beta) using DWT.
• Feature extraction via relative power and variance calculation using CSP.
• Classification using Extreme Learning Machine (ELM).

Main Results:

• Drowsy state exhibited higher theta, alpha, and beta wave activity compared to the awake state.
• Initial ELM accuracy was below 87%.
• CSP combined with ELM achieved high accuracy (91.67%-93.75%) even with limited training data.

Conclusions:

• CSP combined with ELM significantly enhances drowsiness detection accuracy.
• This method reduces training/calibration time while maintaining high classification performance.
• The developed EEG-based system shows promise for preventing drowsy driving risks.