Detection of Changes in Team Dynamics in Simulated Air Battle Management: Univariate vs. Multivariate Analyses

In the future, artificial intelligence (AI) systems may be employed to support team training and these systems will need to be sensitive to team responses to training events. Importantly, team responses may be indexed by changes in team dynamics, which are also related to team performance and adaptability. In this paper, we compare changes in team dynamics assessed via team-level physio-behavioral coupling (PBC) calculated using nonlinear heart rate (HR) variability metrics, complexity of communication patterns, and multivariate combinations of PBC and communication complexity for sensitivity to team responses to critical events. Our data was collected from qualified Air Battle Managers (ABMs) taking part in a simulated training exercise. Each mission of the exercise contained planned and unplanned events that required adaptive responses by the ABMs. For each mission, multidimensional recurrence quantification analysis (MdRQA) was conducted on ABM HR data and the complexity of ABM communications were analyzed using sample entropy and Shannon entropy. Change point detection was conducted on these signals to estimate changes in team dynamics, first in univariate analyses and then on multivariate combinations. Change point times were then compared to times from mission events annotated by a subject matter expert. Analyses showed that Shannon entropy of communication patterns had the worst relative performance of all signals and that multivariate analyses of MdRQA HR and sample entropy of communications performed better than did either variable alone. Our results support the use of dynamical systems measures by future AI trainers to assess team adaptation to training events.