Cognitive Workload Estimation Using LSTM and Ensemble Learning on Multimodal Physiological Data

In this paper, we present a novel cognitive work-load estimation ensemble model, trained and evaluated using data from pilot flying tasks. Our objective is to achieve high and robust estimation accuracy of pilots' cognitive workloads. Remarkably, our proposed model surpasses the performance of the best model in CogPilot Data Challenge 2.0, despite utilizing relatively fewer physiological sensors. Key aspects of our research include integrating modalities obtained at different frequencies and combining time-sequenced raw data besides features. We utilize Long Short-Term Memory networks for time series data estimation and employ ensemble learning to leverage the strengths of multiple models, thereby overcoming their individual limitations.