Continuous Assessment of Mental Workload During Complex Human–Machine Interaction: Inferring Cognitive State from Signals External to the Operator

TL;DR

This paper introduces a machine-learning model to estimate mental workload in helicopter pilots using operational data, which outperforms physiological signals.

Contribution

A novel machine-learning model for real-time mental workload estimation using operational data in complex human-machine interactions.

Findings

The machine-learning model achieved high performance metrics (ROC AUC score 0.836, F1 score 0.842).

Operational data outperformed physiological signals in predicting mental workload.

Abstract

The use of complex human–machine interfaces (HMIs) has grown rapidly over the last few decades in both industrial and personal contexts. Now more than ever, the study of mental workload (MWL) in HMI operators appears essential: when mental demand exceeds task load, cognitive overload arises, increasing the risk of work-related fatigue or accidents. In this paper, we propose a data-driven approach for the continuous estimation of the MWL of professional helicopter pilots in realistic simulated flights. Physiological and operational parameters were used to train a novel machine-learning model of MWL. Our algorithm achieves good performance (ROC AUC score 0.836 ± 0.081, the maximum F1 score 0.842 ± 0.078 and PR AUC score 0.820 ± 0.097) and shows that the operational information outperforms the physiological signals in terms of predictive power for MWL. Our results pave the way towards intelligent systems able to monitor the MWL of HMI operators in real time and question the relevancy of physiology-derived metrics for this task.

Figures