ALTERNATIVE APPROACHES TO IMPROVE PHYSIOLOGICAL PREDICTIONS

Recent advancements in technology have resulted in new biosensors and information processing capabilities that permit on-line, real-time measurement of physiological variables. This has, in turn, given rise to the possibility of developing soldier-specific, data-driven predictive models for assessing physiological status in the battlefield. This paper explores how the accuracy of a predictive model based on first principles physiology can be enhanced by data-driven "black box" techniques of modeling and predicting human physiological variables. Such hybrid techniques are employed here in the prediction of core temperature. Preliminary results show that the mean square error of prediction can be reduced by up to fifty percent for prediction horizons of up to 30 minutes.