EEG brain signals to detect the sleep health of a driver: An automated framework system based on deep learning

Ettahiri, Halima; Ferrández Vicente, José Manuel; Fechtali, Taoufiq · 2022 · Crossref

DOI: 10.3389/fnhum.2022.915276

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 challenge of detecting driver mental fatigue, a critical safety issue, by developing an automated framework using deep learning to classify Electroencephalography (EEG) signals. Traditional methods for distinguishing between normal and sleepy states, such as Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and K-Nearest Neighbors (KNN), typically achieve low accuracy (ranging from 64% to 88%) and rely on hand-designed feature extraction. These conventional approaches struggle with the limited availability of labeled EEG data and the complexity of neurological disorganization associated with fatigue. The authors propose a Convolutional Neural Network (CNN) model to automate feature learning and improve classification precision, specifically addressing the constraint of small dataset sizes through data augmentation techniques. The experimental design involved 20 healthy volunteers (aged 25–36) divided into normal sleep and sleep deprivation groups. EEG signals were recorded using an eight-channel wearable cap (ENOBIO 8) at a sampling frequency of 500 Hz, filtered between 1–30 Hz to remove noise. Due to the small dataset size, the researchers implemented a specific data augmentation strategy involving the multiplication of signal samples by Gaussian noise arrays rather than geometric transformations, which would distort the temporal nature of EEG data. The proposed model is a 1D CNN architecture featuring three convolutional layers with kernel sizes of 5×5, 3×3, and 3×3, followed by fully connected layers with 4,096 neurons each. The network utilized ReLU activation functions and Softmax output for binary classification. Training employed transfer learning and fine-tuning, with 80% of the data used for training and 20% for validation and testing. Model performance was evaluated using 10-fold cross-validation. The results demonstrate that the proposed CNN framework significantly outperforms traditional machine learning classifiers. The deep learning model achieved a classification accuracy of 97.3%, compared to maximum accuracies of 88.7% for SVM, 87.5% for LDA, and 86.25% for KNN when using standard feature extraction methods. Detailed metrics for the CNN model showed an accuracy of 0.97 for the normal class and 0.89 for the sleepy class, with precision scores of 0.91 and 0.85, respectively. The validation loss decreased effectively during training, indicating robust convergence. The study confirms that CNNs can effectively learn discriminative features directly from raw EEG signals without manual feature engineering, even with limited data when augmented appropriately. The significance of this work lies in its demonstration that deep learning can provide a highly accurate, automated solution for driver fatigue detection, surpassing the limitations of traditional classifiers. By achieving near-perfect accuracy with a minimal number of electrodes and a small dataset, the framework offers a viable path for real-time, non-intrusive monitoring systems in vehicles. This approach reduces the reliance on expert neurologists for signal interpretation and addresses the data scarcity problem common in biomedical signal processing, suggesting that CNN-based models are superior for binary classification tasks in cognitive neuroscience applications.

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
• Methodological Resource: tool software