Driver Drowsiness EEG Detection Based on Tree Federated Learning and Interpretable Network.

Area of Science:

• Neuroscience
• Machine Learning
• Traffic Safety

Background:

• Driver drowsiness is a major cause of traffic accidents.
• Electroencephalogram (EEG) signals are crucial for identifying drowsiness but are often stored as small, distributed datasets.
• Existing methods struggle with privacy and efficient data integration.

Purpose of the Study:

• To develop an efficient and accurate privacy-preserving system for monitoring driver drowsiness.
• To propose a novel fusion model combining tree Federated Learning (FL) and Convolutional Neural Networks (CNN).
• To enable identification and explanation of drowsiness states while protecting user privacy.

Main Methods:

• A fusion model integrating tree Federated Learning (FL) with Convolutional Neural Networks (CNN) was developed.
• Each client utilized a CNN with a Global Average Pooling (GAP) layer, sharing model parameters.
• Tree FL structured communication as a graph, enabling parallel parameter transmission; Class Activation Mapping (CAM) identified key EEG features.

Main Results:

• The proposed method achieved higher average accuracy (73.56%), F1-score (73.26%), and AUC (78.23%) compared to traditional classification methods.
• Demonstrated superior privacy protection compared to traditional FL algorithms.
• Showcased improved communication efficiency in the drowsiness monitoring system.

Conclusions:

• The tree FL-CNN fusion model offers an effective solution for privacy-preserving driver drowsiness detection using EEG.
• The system enhances communication efficiency and privacy protection in federated learning environments.
• This approach holds significant potential for reducing traffic accidents caused by driver fatigue.