Identification of Potential Task Shedding Events Using Brain Activity Data

In Human–Machine Teaming environments, it is important to identify potential performance drops due to cognitive overload. If identified correctly, they can help improve the performance of the human–machine system by offloading some tasks to less cognitively overloaded users. This can help prevent user error that can result in critical failures. Also, it can improve productivity by keeping the human operators at an optimal performance state. This paper explores a new method for identifying user cognitive load by a three-class classification using brain activity data and by applying a convolutional neural network and long short-term memory model. The data collected from a set of cognitive benchmark experiments were used to train the model, which was then tested on two separate datasets consisting of more ecologically valid task environments. We experimented with various models built with different benchmark tasks to explore which benchmark tasks were better suited for the prediction of task shedding events in these compound tasks that are more representative of real-world scenarios. We also show that this method can be extended across-tasks and across-subject pools.