The Department of Defense plans to spend $1.7 billion over the next five years to stand up a new Joint Artificial Intelligence Center with goals to develop strategic plans, adopt and transition artificial intelligence, machine learning and emerging technologies into operational use (Longwell, 2018). Until roadmaps have matured, it is unclear how much of that budget will go towards training.

Several commercial industries have implemented solutions using data analytics to improve operations. Many in the military training industry are beginning to design architectures and plan research studies using learning analytics to predict performance, personalize and adapt training to optimize human performance. Large amounts of training data sets for effectively training the networks is one of the biggest challenges. Few researchers have assessed machine learning solutions that include physiological metrics to adapt learning. Collins Aerospace capitalizes on the unique data collection from previous, privately funded research conducted over the past two years. This research collected data on 30 pilots flying multiple flight maneuvers in a simulator and in a live aircraft with over 50 plus hours of live flight time. We collected metrics on task performance and cognitive workload.

Collins developed several deep neural networks to predict future states of student performance. We compared the results of our predicted performance using only task performance measures with the results of task performance measures in combination with cognitive workload metrics. The results show when cognitive workload is included in our deep neural networks, it increased the performance prediction to an extremely high level of accuracy.

Joseph Williams

Affiliation

Refine Results

The Value of Cognitive Workload in Machine Learning Predictive Analytics