Recognizing and explaining driving stress using a Shapley additive explanation model by fusing EEG and behavior...

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Transportation Safety

Background:

• Driving stress is a significant contributor to road traffic accidents.
• Existing driving stress recognition models often lack interpretability, hindering practical application.
• Understanding the interplay between brain activity and driving behavior is crucial for safety.

Purpose of the Study:

• To develop an explainable framework for driving stress recognition using electroencephalography (EEG) and behavior data.
• To improve the accuracy of stress recognition by fusing multimodal data.
• To analyze the dependency effects between brain activity, driving behavior, and stress levels.

Main Methods:

• Feature extraction and selection from EEG and behavior datasets.
• Application of seven machine learning algorithms for stress level identification (low, medium, high).
• Utilizing SHapley Additive exPlanation (SHAP) for model interpretability and decision trees for fuzzy rule extraction.

Main Results:

• Fusion of EEG and behavior features significantly improved accuracy by 8.56% (vs. EEG alone) and 26.51% (vs. behavior alone), achieving 84.93% overall accuracy.
• SHAP analysis and decision trees revealed key physiological and behavioral indicators of stress, including reduced speed, lane deviation, and altered EEG power in θ and β bands.
• Identified higher synchronicity in brain region activity within the same frequency band under stress.

Conclusions:

• The proposed explainable framework effectively quantifies driving stress by integrating EEG and behavior data.
• Multimodal data fusion offers superior accuracy in driving stress recognition compared to single-modality approaches.
• The study provides insights into the physiological and behavioral manifestations of driving stress, aiding in the development of safer driving systems.