Beyond Dyads: Machine Learning for Team Performance Prediction in Multi-Member Teams

Team operations, in which multiple members collaborate to accomplish tasks together, play a pivotal role in many safety-critical domains, such as military training, naval operations, and medical care. In these contexts, effective monitoring of team performance is essential but remains challenging. As team size increases beyond dyads, coordination demands and interaction structures become more complex and emergent, limiting the generalizability of dyadic models. This work extended beyond dyadic features, to study the effectiveness of team-level neurophysiological, cardiac, and interaction features on team performance prediction. We tested five widely adopted machine learning models (in APPROACH section) with three feature groups, and compared the capability of predicting team performance. Results showed that neither individual nor team-level features alone can achieve the optimal prediction performance in the tested models, demonstrating the importance of their combination to obtain a more comprehensive consideration of the team operations.