EEG-Based Driving Fatigue Detection Using a Two-Level Learning Hierarchy Radial Basis Function

Ren, Ziwu; Li, Rihui; Chen, Bin; Zhang, Hongmiao; Ma, Yuliang; Wang, Chushan; Lin, Ying; Zhang, Yingchun · 2021 · Crossref

DOI: 10.3389/fnbot.2021.618408

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 accurately detecting driving fatigue using electroencephalography (EEG) signals, a critical issue given the high risk of traffic accidents caused by drowsy driving. While EEG is a promising non-invasive modality, extracting informative features from noisy signals remains difficult. The authors propose a Two-Level Learning Hierarchy Radial Basis Function (RBF-TLLH) network to enhance classification accuracy and efficiency. Traditional RBF networks rely heavily on optimizing numerous parameters, which often leads to high computational costs and slow convergence. The proposed RBF-TLLH method simplifies this by requiring the optimization of only three core parameters, thereby improving reliability and applicability. The experimental design involved six healthy male participants who underwent simulated driving sessions to collect EEG data in both alert and fatigue states. To induce fatigue, participants slept for only four hours prior to the experiment and drove on a monotonous road until drowsy signs appeared. EEG signals were recorded using a 32-channel system at 500 Hz. Data preprocessing included downsampling to 200 Hz and band-pass filtering (1–45 Hz). The signals were segmented into 10-second windows, resulting in 1,440 total samples. Principal Component Analysis (PCA) was employed for dimensionality reduction, retaining the first 10 principal components which accounted for over 80% of the variance. The classification model utilized a two-level hierarchy: the lower level used a Regularized Orthogonal Least Squares combined with D-optimality (ROLS+D-opt) algorithm to determine network weights, while the upper level used Particle Swarm Optimization (PSO) to globally optimize the three key parameters: basis function width, regularization parameter, and D-optimality weighting parameter. The results demonstrated that the RBF-TLLH approach achieved a mean classification accuracy of 92.71% and an Area Under the Receiver Operating Characteristic curve (AUC) of 0.9199. This performance significantly outperformed other widely used artificial neural networks, including Back-Propagation ANN (ANN-BP), PSO-optimized ANN (ANN-PSO), and standard RBF networks using only the ROLS+D-opt algorithm. Statistical analysis via paired t-tests confirmed that RBF-TLLH significantly outperformed the ANN classifiers. Furthermore, the RBF-TLLH model exhibited higher sensitivity and lower variance compared to the standard RBF-ROLS+D-opt model, indicating superior robustness in detecting fatigue states. The computational complexity of the TLLH scheme was also noted to be significantly lower than methods optimizing all network parameters directly. The significance of this work lies in demonstrating that the RBF-TLLH framework provides a reliable, efficient, and accurate method for EEG-based driving fatigue detection. By reducing the number of parameters requiring global optimization, the method mitigates the computational burdens associated with traditional RBF networks while maintaining high classification performance. These findings suggest that the proposed approach is a promising tool for developing real-time, non-invasive driver monitoring systems to enhance road 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
• Theoretical Contribution: computational model