Identification of Adaptive Thermogenesis in Humans Using Machine Learning from CALERIE Dataset

Abstract

In recent years, weight loss has become a serious concern in humans due to a rise in obesity rates. Different solutions have been developed for different body types. Nonetheless, maintaining a weight loss is a challenge since most people end up regaining the lost weight. The phenomenon is known as adaptive thermogenesis (AT). Using the CALERIE study dataset, the article proposes to use machine learning to estimate the chance of adaptive thermogenesis. Adaptive thermogenesis is determined by the difference between predicted and measured Resting metabolic rates as provided by the CALERIE study dataset. Depending on whether or not AT is below 5 percent of RMR, AT is scaled into binary form. Three state of-the-art machine learning algorithms were then applied to the dataset, namely logistic regression, decision tree classifiers, and explainable boosting machines. These models explain AT in humans in the simplest form possible. Different training rates led to different results for each algorithm. Despite the explainable boosting machines' higher accuracy when given more data for training, the logistic regression classifier provided better generalization as training data reduced from 80% to 60%. Several modelling approaches provided results that underlined that body temperature, BMI, alcohol intake, and waist to height ratio (WhtR) were more relevant with AT. This explains why thermogenesis is observed in warm-blooded animals, an occurrence that is primarily determined by body temperature. AT is increased in humans with a poor body type and excessive alcohol consumption. In other words, a poor lifestyle promotes AT in humans