Detection and analysis of driver fatigue stages with EEG signals

Demir, Ahmet; Bekiryazıcı, Şule; Coşkun, Oğuzhan; Eken, Recep; Yılmaz, Güneş · 2022 · Crossref

DOI: 10.5505/pajes.2022.89327

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the critical safety issue of driver fatigue, a primary cause of traffic accidents, by developing a system to detect and analyze fatigue stages using electroencefalography (EEG) signals. While various methods exist for fatigue detection, such as eye tracking or heart rate monitoring, they often suffer from low accuracy and applicability. EEG is identified as a more reliable method because it directly measures neurophysiological activity. The research aims to classify driver fatigue into distinct stages using a dataset from National Chiao Tung University, which records EEG signals during sustained-attention driving tasks. The experimental design utilized EEG data from 27 participants who underwent a 90-minute driving simulation in a virtual reality environment. The dataset was divided into four classes representing progressive fatigue levels: Class 1 (0–22.5 minutes, awake), Class 2 (22.5–45 minutes, intermediate), Class 3 (45–67.5 minutes, drowsy signs), and Class 4 (67.5–90 minutes, struggling to stay awake). EEG signals were filtered into five frequency bands: Delta, Theta, Alpha, Beta, and Gamma. Feature extraction involved applying Fourier and Wavelet transforms to derive seven specific attributes, including standard deviation, mean, variance, skewness, and kurtosis. To manage the high dimensionality of the resulting feature matrix, Principal Component Analysis (PCA) was employed for dimensionality reduction. For classification, the study implemented a Divide and Conquer algorithm, which decomposes the multi-class problem into binary sub-problems. Three sub-classifiers were evaluated: k-Nearest Neighbors (k-NN), Support Vector Machines (SVM), and Linear Discriminant Analysis (LDA). The system’s performance was assessed using 10-fold cross-validation, total classification success, and error matrices. The results indicated that SVM generally outperformed the other algorithms. The average classification success rates were 87.9% for k-NN, 88.5% for SVM, and 81.6% for LDA. Specifically, the highest classification accuracy of 93.2% was achieved using the SVM classifier for the fourth fatigue level (67.5–90 minutes). Individual frequency band analyses showed that Delta and Theta bands yielded higher classification accuracies compared to Beta and Gamma bands across all classifiers. The significance of this research lies in demonstrating the efficacy of EEG-based systems for real-time driver fatigue monitoring. By achieving high classification accuracy, particularly in advanced fatigue stages, the study validates the use of SVM within a Divide and Conquer framework for processing EEG data. The findings suggest that focusing on lower frequency bands (Delta and Theta) and utilizing PCA for feature reduction can enhance the reliability of fatigue detection systems, potentially contributing to improved road safety technologies.

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 qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

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

• drowsiness detection algorithms
• drowsiness

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