A deep learning approach for metabolic rate prediction

Abstract

The sudden increase of wearable devices has led to the generation of an abundance of data. As a result, researchers can use such data to perform analyses and generate recommendations. A crucial factor in the research field of nutrition and dietetics is the accurate measurement of metabolic rate, as it can be used to estimate several other variables, e.g. calorie expenditure. Nonetheless, limited studies have been conducted to examine the use of machine learning models for metabolic rate prediction based on data generated from wearable devices. Therefore, in this paper, a neural network architecture is proposed, able to predict a subject's metabolic rate exploiting only such data. Experimental results demonstrated that the proposed methodology can outperform conventional algorithms in prediction accuracy of real-world data. Furthermore, results indicated that the trivial time taken for the network to predict the metabolic rate makes it suitable for wearable devices deployment.