Clustering for mitigating subject variability in driving fatigue classification using electroencephalography...

Area of Science:

• Neuroscience
• Machine Learning
• Transportation Safety

Background:

• Electroencephalography (EEG)-based driver fatigue classification models show promise but face challenges in real-world application due to individual signal variability.
• Developing universal models is difficult, often requiring impractical retraining with new subject data, especially for fatigue states.

Purpose of the Study:

• To address the limitations of current EEG-based driver fatigue detection systems.
• To propose a hybrid clustering and classification approach for more adaptable and effective fatigue detection.

Main Methods:

• Employed unsupervised clustering to group subjects based on EEG functional connectivity (FC) in an alert state.
• Applied classification models to each identified cluster for predicting alert and fatigue states.

Main Results:

• Classification accuracy improved when applied to clustered subjects compared to non-clustered scenarios.
• Clustering successfully grouped subjects with similar FC characteristics, enhancing the classification performance.

Conclusions:

• The hybrid clustering-classification method offers a practical and realistic solution for driver fatigue detection.
• This approach improves the adaptability and effectiveness of EEG-based fatigue detection systems in real-world settings.