Investigating the factors affecting obesity using machine learning algorithms

Obesity is a multifactorial public health challenge, influenced by a complex interplay of behavioral, dietary, genetic, and lifestyle factors. Traditional statistical methods often fall short in capturing nonlinear relationships and high-dimensional interactions within such data. This study aims to identify the most influential predictors of obesity using four machine learning-based feature selection methods, thereby offering robust insights for public health interventions and policy design. A dataset comprising 2,111 records from individuals in Mexico, Peru, and Colombia—partially augmented using the SMOTE technique—was analyzed using Boruta, Recursive Feature Elimination (RFE), Lasso Logistic Regression, and Genetic Algorithms. Variables included demographic, behavioral, dietary, and physical activity-related features. All analyses were conducted in R. Across all four methods, high-calorie food consumption, frequent snacking, low water intake, reduced physical activity, and family history of overweight were consistently identified as key predictors of obesity. In contrast, variables such as gender, smoking, and transportation mode were not selected by any method, suggesting limited predictive value in the given context. Some features, like alcohol intake and vegetable consumption, showed algorithm-specific relevance. The convergence of findings across multiple machine learning algorithms strengthens the validity of selected predictors, emphasizing the role of lifestyle and dietary habits in obesity risk. The study highlights the utility of multi-algorithmic feature selection in deriving interpretable and reliable insights from complex health data, with implications for designing targeted intervention strategies.