Analysis of Alpha and Theta Band to Detect Driver Drowsiness Using Electroencephalogram (EEG) Signals

Sivakumar, Pradeep Kumar; Selvaraj, Jerritta; Ramaraj, Krishnakumar; Sahayadhas, Arun · 2021 · Crossref

DOI: 10.34028/18/4/10

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical safety issue of driver drowsiness, a leading cause of road accidents, by developing a detection system based on Electroencephalogram (EEG) signals. The study focuses specifically on analyzing Alpha (8–13 Hz) and Theta (4–8 Hz) frequency bands, which are known to exhibit significant changes during drowsiness. The primary objective is to identify specific EEG channels and features that can reliably distinguish between alert and drowsy states to enable real-time driver alerting. The experimental design involved ten healthy male participants (ages 19–32) who engaged in a simulated driving task using a "Speed Dreams" game at a constant speed of 70 km/h. EEG data was recorded from 21 channels using an Allengers Polysomnography device at a sampling rate of 256 Hz. To induce drowsiness, experiments were conducted during circadian low points (midnight, early morning, and afternoon). Raw EEG signals were pre-processed using a 6th-order Butterworth band-pass filter (0.5–49 Hz) to remove artifacts. The signals were then decomposed into sub-bands via Fast Fourier Transform. Eleven linear and non-linear features, including Mean, Kurtosis, and Hurst exponent, were extracted from the Alpha and Theta bands. Statistical significance was determined using Analysis of Variance (ANOVA), with only features showing p<0.05 retained for classification. The alert and drowsy states were classified using Quadratic Discriminant Analysis (QDA), Linear Discriminant Analysis (LDA), and K-Nearest Neighbour (KNN) algorithms, validated with a 75/25 train/test split. The results indicate that the Hurst exponent and Kurtosis are the most statistically significant features for both Alpha and Theta bands. In the Alpha band, the Hurst feature achieved a maximum classification accuracy of 92.86% using the KNN classifier on channels F8 and T6, while the Kurtosis feature achieved 100% accuracy on channel T5. For the Theta band, the Hurst feature achieved 100% accuracy on channel F8, and Kurtosis achieved a maximum of 92.85% on channels FP1, CZ, and O1. A comparative analysis using the KNN classifier revealed that the Alpha band generally outperformed the Theta band for the Hurst feature, while the Theta band showed higher accuracy for the Kurtosis feature in most channels. The study identified a subset of 12 channels across frontal, temporal, and occipital regions as optimal for drowsiness detection, suggesting that channel reduction is feasible to improve driver comfort. The significance of this work lies in its demonstration that specific EEG channels and non-linear features can effectively detect drowsiness with high accuracy. By identifying that fewer than 21 channels are necessary for robust classification, the research supports the development of more comfortable, wearable EEG-based monitoring systems. The findings provide a foundation for integrating such systems into real-time vehicle safety applications, although the authors note that future work must validate these results with larger, more diverse participant groups in real-time driving conditions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• drowsiness
• microsleep

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data