Classification of a Driver's cognitive workload levels using artificial neural network on ECG signals

Area of Science:

• Cardiovascular Physiology
• Computational Neuroscience
• Human Factors Engineering

Background:

• Driver cognitive workload assessment is crucial for road safety.
• Electrocardiography (ECG) offers a non-invasive method to infer physiological states.
• Existing methods for workload classification often lack real-time applicability.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for classifying driver cognitive workload.
• To utilize electrocardiography (ECG) signals as the primary data source for workload classification.
• To investigate the efficacy of specific ECG time-domain and frequency-domain measures.

Main Methods:

• ECG signals were recorded from 15 male participants during a simulated driving task with varying cognitive loads.
• Key time-domain (mean IBI, SD IBI, RMSSD) and frequency-domain (LF, HF, LF/HF ratio) ECG measures were extracted.
• A three-step data processing procedure involving measure selection, workload level definition, and normalization was applied.
• A feed-forward ANN with scaled conjugate gradient back-propagation was implemented for classification.

Main Results:

• The ANN model demonstrated high accuracy on learning data (95%) and satisfactory performance on unseen testing data (82%).
• The selected ECG measures, after normalization, effectively captured variations in cognitive workload.
• The model successfully differentiated between low, medium, and high cognitive workload levels.

Conclusions:

• ANN models can effectively classify driver cognitive workload using ECG signals.
• The developed method provides a promising approach for real-time, non-invasive workload monitoring in drivers.
• Further research can explore broader participant demographics and more complex driving scenarios.