Artificial Neural Networks in the Determination of the Fluid Intake Needs of Endurance Athletes

Abstract

The aim of this study was to assess the efficacy of using artificial neural networks (ANNs) to classify hydration status and predict the fluid requirements of endurance athletes. Hydration classification models were built using a total of 237 data sets obtained from 148 participants (106 males,42 females) in field-and laboratory studies involving running or cycling. 116 data sets obtained from athletes who completed endurance events euhydrated (plasma osmolality: 275-295 mmol.kg^-1) following ad libitum replenishment of fluid intake was used to design prediction models. A filtering algorithm was used to determine the optimal inputs to the models from a selection of 13 anthropometric, exercise performance, fluid intake and environmental factors. The combination of gender, body mass, exercise intensity and environmental stress index in the prediction model generated a root mean square error of 0.24 L.h^-1 and a correlation of 0.90 between predicted and actual drinking rates of the euhydrated participants. Additional inclusion of actual fluid intake resulted in the design of a model that was 89% accurate in classifying the post-exercise hydration status of athletes. These findings suggest that the ANN modelling technique has merit in the prediction of fluid requirements and as a supplement to ad libitum fluid intake practices.