Classifying metro drivers' cognitive distractions during manual operations using machine learning and random...

Area of Science:

• Human-Computer Interaction
• Transportation Safety
• Biomedical Engineering

Background:

• Cognitive distractions pose a significant risk to metro driver safety, often lacking observable behavioral indicators.
• Existing methods struggle to identify subtle cognitive distractions in real-time during manual driving tasks.

Purpose of the Study:

• To develop a non-invasive method for identifying cognitive distractions in metro drivers using Electrocardiogram (ECG) signals.
• To differentiate between the presence and severity of cognitive distractions during simulated metro driving.

Main Methods:

• Collected ECG data from metro drivers during simulated driving experiments using wearable devices.
• Extracted ultra-short-term heart rate and heart rate variability (HR-HRV) features using 30-s and 60-s time windows.
• Employed machine learning algorithms, including Random Forest-Recursive Feature Elimination (RF-RFE), Decision Trees (DT), and XGBoost, for classification.

Main Results:

• Optimal models identified: DT with one HR-HRV feature (30-s window) for binary classification and XGBoost with 20 HR-HRV features (60-s window) for multi-class classification of driving distractions.
• Key HR-HRV features associated with distractions include NN20, pNN20, SD1/SD2, Max-HR, Min-HR, and MEDNN.
• Detection of cognitive distractions during the parking phase using HR-HRV features proved challenging.

Conclusions:

• ECG-derived HR-HRV features offer a promising approach for detecting cognitive distractions in metro drivers during operation.
• Machine learning models, particularly DT and XGBoost, can effectively classify distraction presence and severity based on physiological signals.
• Further research is needed to enhance the detection of distractions in non-driving phases like parking.