Assessment of Scalability and Adaptability in Federated Learning Framework for Robust Driver Stress Monitoring.

Area of Science:

• Computational neuroscience
• Machine learning
• Biomedical engineering

Background:

• Driving requires constant attention, but stress from external factors impairs performance.
• Centralized machine learning for driver stress detection raises privacy concerns regarding physiological data.
• Federated learning (FL) offers a privacy-preserving approach for collaborative model training.

Purpose of the Study:

• To evaluate the feasibility of using federated learning for driver stress detection.
• To develop a privacy-preserving machine learning framework for real-time stress monitoring.
• To assess the performance of FL models using multi-modal physiological signals.

Main Methods:

• Collected multi-modal physiological signals (ECG, respiration, accelerometer) from 20 drivers in various driving scenarios.
• Implemented a decentralized federated learning framework with an improved ResNet model.
• Evaluated model performance using subject-dependent and subject-independent learning paradigms.

Main Results:

• Federated learning models achieved high accuracy in detecting driver stress states.
• Subject-independent FL models demonstrated strong generalization across unseen subjects, reaching 97.82% accuracy.
• FL models provided superior privacy protection and scalability compared to centralized approaches.

Conclusions:

• Federated learning is a feasible and effective method for privacy-preserving driver stress detection.
• Subject-independent FL models show significant potential for real-world deployment in vehicular and biomedical applications.
• The proposed framework supports real-time and clinical stress monitoring systems with robust adaptability.