Real-time cognitive workload assessment using non-intrusive methods: a systematic review
综述了50项同行评审研究，分析实时认知负荷评估中生理和行为监测的方法趋势，发现可穿戴设备使用占主导（约74%），传统机器学习（32%）和统计模型（约21%）比深度学习（约14%）更常用，为任务需求下的测量和模型选择提供指南。
Real-time cognitive workload (CWL) assessment has been used to enhance human performance and safety across various operational domains. This review synthesizes findings from 50 peer-reviewed studies to examine current practices, methodological trends, and technological advances in physiological and behavioural CWL monitoring. Studies utilized electrocardiography (ECG), photoplethysmography (PPG), electrodermal activity (EDA), eye-tracking, electroencephalography (EEG), and skin temperature (SKT). The use of wearable devices were predominant (∼74%). Task categorization into cognitive, perceptual, motor, and physical domains revealed alignment between physiological measures and task demands. Computational approaches favour traditional machine learning (32%) and statistical models (∼21%) over advanced deep learning models (∼14%). The use of hybrid approaches (∼22%), where multiple models are combined or applied in parallel rather than using a single model, suggests evolution towards adaptive frameworks for real-world implementation. The review offers guidelines on measurement and model selection based on task requirements and outlines future directions for real-world CWL system deployment.