Effectiveness of Physiological and Psychological Features to Estimate Helicopter Pilots' Workload: A Bayesian Network Approach

• Besson Patricia
• Bourdin Christophe
• Bringoux Lionel
• Dousset Erick
• Maiano Christophe
• Marqueste Tanguy
• Mestre Daniel R.
• Gaetan Sophie
• Baudry Jean-Pierre
• Vercher Jean-Louis

• Human factors
• Intelligent systems
• Cognitive science
• Graph theory
• Human computer interaction

Despite growing interest over the decades, the question of estimating cognitive workload of operators involved in complex multitask operations, such as helicopter pilots, remains a key issue. One of the main difficulties facing workload inference models is that no single specific indicator of workload exists, so that multiple sources of information have to be inputted to the model. The question then arises as to the nature and the quantity of features to be used for increasing model performance. In this research, done in cooperation with Eurocopter, the effectiveness of physiological, psychological, and cognitive features for estimating helicopter pilots' workload was systematically investigated, using Bayesian networks (BNs). The study took place in two different contexts: a constrained laboratory situation with low ecological validity and a more realistic and challenging situation relying on virtual reality. The constrained conditions of the laboratory study allowed us for testing various combinations of entropy-based physiological, cognitive, and affect features as inputs of BN models. These three different kinds of features are shown to carry complementary information that can be used with advantage by the model. The results also suggest that increasing the number of physiological inputs improves the model performance. The second study aimed at challenging some of these conclusions in a more ecological context, by using the NH90 full-flight simulator of the Helisim company. The results emphasize the problem of accessing the ground truth, as well as the need for an efficient feature selection or extraction step prior to the classification step.