Early obesity risk prediction via non-dietary lifestyle factors using machine learning approaches.

Abstract

Obesity poses a significant health threat, contributing to the development of noncommunicable diseases (NCDs). Early identification of individuals at higher risk for obesity is crucial for implementing effective prevention strategies. This study explores the viability of non-dietary factors such as lifestyle, family history, and demographics as predictors of obesity risk. The dataset comprised 1068 males and 1043 females, aged between 14 and 61 years. Only non-dietary factors were used to build the machine learning models, including decision tree, random forest, support vector classification (SVC), k-nearest neighbour (KNN), and Gaussian Naïve Bayes (GNB). Random forest emerged as the optimal model, demonstrating 66.9% test accuracy, 66.4% precision, 66.9% recall, 66.4% F1-score, 94.5% specificity and 92.3% area under the receiver operating characteristic curve (AUC-ROC). Variability of the models' performance was also evaluated through bootstrapping. Lifestyle factors, while less impactful than family history and demographics, also contributed to predictive power. This indicates the potential for predicting obesity while relying less on dietary data, paving the way for future studies to refine predictive models. This could play a crucial role in identifying lifestyle factors as predictors of obesity, thereby preventing and intervening early to address obesity-related complications.