EEG-based driver fatigue detection using hybrid deep generic model

San, Phyo Phyo; Ling, Sai Ho; Chai, Rifai; Tran, Yvonne; Craig, Ashley; Nguyen, Hung · 2016 · Crossref

DOI: 10.1109/embc.2016.7590822

archive: archived pipeline: cataloged verified

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 fatigue, a major contributor to traffic accidents worldwide, by proposing a novel method for detecting fatigue states using electroencephalography (EEG) signals. While EEG is a reliable measure of neurophysiological activity, traditional classification methods often struggle with the high-dimensional and non-linear nature of the data. The authors aim to overcome the limitations of existing deep learning models, which may not be optimal for classification due to parameter placement, and Support Vector Machines (SVM), which lack the ability to learn complex invariances independently. To this end, they introduce a hybrid Deep Generic Model-based SVM (DGM-SVM) framework that integrates unsupervised high-level feature extraction with supervised classification. The methodology involves a two-stage process. First, raw EEG data is preprocessed to remove artifacts such as ocular and muscular activities using Second Order Blind Identification (SOBI) and canonical correlation. The cleaned data is then segmented into 20-second windows, which are further divided into 2-second overlapping units. The core of the system is a Deep Belief Network (DBN) constructed by stacking Restricted Boltzmann Machines (RBMs). This deep architecture is trained in a greedy, layer-wise manner to extract high-level, invariant features from the raw, high-dimensional EEG inputs without manual feature engineering. These extracted features are then fed into an SVM classifier, which utilizes a polynomial kernel to determine the driver’s status as either fatigued or alert. The performance is evaluated using sensitivity, specificity, and accuracy metrics. The experimental validation utilized EEG data from five healthy participants engaged in a driving simulation task. Data was recorded from 32 channels at a sampling rate of 256 Hz. The dataset was divided into alert states (first 5 minutes of driving) and fatigue states (last 5 minutes before stopping). The final dataset comprised 730 units, split into training (438 units from three participants) and testing (292 units from two participants) sets. The optimized DGM-SVM model, featuring 32 input units and 23 hidden neurons in the RBM layer, achieved a testing accuracy of 73.29%, with a sensitivity of 91.10% and a specificity of 55.48%. In comparison, a traditional SVM classifier using Power Spectrum Density (PSD) features achieved only 61.00% accuracy, 70.00% sensitivity, and 52.00% specificity. The DGM approach extracted significantly fewer features (23) compared to PSD (121) while delivering superior classification performance. The study concludes that the hybrid DGM-SVM framework is an effective and robust method for driver fatigue detection. By consolidating unsupervised feature extraction via deep hierarchical RBMs with SVM classification, the system enhances both the quality of feature representation and the accuracy of the final decision boundary. The results demonstrate that deep learning models can successfully handle raw, high-dimensional EEG data to identify fatigue states with high sensitivity, offering a promising advancement for real-time driver monitoring systems in biomedical engineering and transportation safety.

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

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