Multi-model Approach to Human Functional State Estimation

Abstract

Listen

Translate article

With the growth and affordability of the wearable sensors market, there is increasing interest in leveraging physiological signals to measure human functional states. However, the desire to produce a reliable universal classifier of functional state assessment has proved to be elusive. In efforts to improve accuracy, we theorize the fusion of multiple models into a single estimate of human functional state could outperform a single model operating in isolation. In this paper, we explore the feasibility of this concept using a workload model development effort conducted for an Unmanned Aircraft System (UAS) task environment at the Air Force Research Laboratory (AFRL). Real-time workload classifiers were trained with single-model and multi-model approaches using physiological data inputs paired with and without contextual data inputs. Following the evaluation of each classifier using two model evaluation metrics, we conclude that a multi-model approach greatly improved the ability to reliably measure real-time cognitive workload in our UAS operations test case.