One size does not fit all: a support vector machine exploration of multiclass cognitive state classifications using...

Area of Science:

• Machine Learning
• Human Factors Engineering
• Physiological Computing

Background:

• Cognitive workload (CWL) assessment is crucial for operator performance and safety in demanding fields like aviation.
• Traditional CWL assessment methods are often subjective or lack real-time applicability.
• Machine learning offers a dynamic approach to CWL prediction using physiological data.

Purpose of the Study:

• Develop and evaluate support vector machine (SVM) models for predicting CWL from physiological data.
• Create binary and multiclass SVM classifiers for nuanced CWL prediction.
• Investigate the benefits of individualized models for improved CWL prediction accuracy.

Main Methods:

• Collected physiological data (ECG, pupillometry) from participants in an aviation simulator.
• Trained binary and multiclass SVM models on task demand and subjective CWL ratings.
• Evaluated individualized and combined-subject models, incorporating feature selection.

Main Results:

• Binary SVMs achieved high accuracy (70.5%-80.4%) in predicting task demand and workload.
• Multiclass models showed good discrimination (AUC-ROC: 0.75-0.79) across different CWL levels.
• Individualized models demonstrated a significant average accuracy improvement of 13% over combined models.

Conclusions:

• SVMs effectively predict CWL using physiological data.
• Tailored, individualized multiclass SVM models offer superior granularity and accuracy for CWL prediction.
• These findings support the development of adaptive automation systems for enhanced safety and performance.