Quantitative analysis of obesity predictors: Evidence from the Ravansar non-communicable disease (RaNCD) cohort study.

Abstract

Obesity is a significant public health issue, influenced by dietary, environmental, socioeconomic, and behavioral factors. Identifying these predictors is crucial for effective prevention strategies and policies. Cohort datasets provide a valuable resource for understanding long-term health trends and risk factors, offering robust insights into obesity determinants. This study aims to identify obesity predictors using cohort data from western Iran. This study used the main phase and first follow-up data from the RaNCD cohort study, including 2064 participants with an obesity incidence of 18 %. We used a hybrid feature selection approach combining filter and wrapper methods using Linear Regression (LR) and Decision Tree (DT) algorithms. Training, testing, and evaluation were performed in Jupyter Notebook using Python 3.12. Fifty-seven variables were identified as significant predictors of BMI. Among them, anthropometric measures such as weight (r = 0.688) and waist circumference (r = 0.686) showed the strongest positive associations, while physical activity (MET, r = -0.151) had the strongest negative correlation. Anthropometric, demographic, and laboratory measures yielded the most accurate BMI predictions, whereas dietary and lifestyle factors contributed less, likely due to limited variability in self-reported data. According to the results, anthropometric, demographic, and lab tests variables were three common predictors for obesity. Integrating these with basic clinical and laboratory data could improve early detection of metabolic risk in primary care screening programs. Public health programs should emphasize culturally tailored approaches to physical activity and dietary behavior to address obesity at the community level.