Workers/crews’ mental workload dynamics in closed cabins: A task-difficulty-adaptive random forest model for instrument monitoring tasks using multimodal biosignals

Abstract

Background

Accurate assessment of operator mental workload (MWL) is critical for ensuring safety in closed-cabin environments, yet traditional contact-based sensors are intrusive.

Objective

This study aimed to develop and validate a fully non-contact, multimodal physiological monitoring framework for assessing levels of Mental Workload in closed-cabin environments.

Methods

This study employed a millimeter-wave radar and a camera to non-contactually acquire ECG, respiration, and eye movement signals from 30 participants performing a four-level monitoring task.

Results

Physiological features demonstrated a significant correlation with task difficulty. A Random Forest classifier built on these features achieved 83.33% accuracy in distinguishing the four MWL levels.

Conclusions

This study validates a fully non-contact, multimodal physiological monitoring framework, providing a practical paradigm for non-intrusive, continuous cognitive state assessment in safety-critical domains.