A comprehensive analysis of BMI prediction using machine learning and biochemical markers: Insights from NHANES data

This study investigates the relationship between obesity and common biochemical markers, focusing on renal function indicators, using machine learning models to diagnose body mass index (BMI) and identify clinically relevant markers associated with it. Data from 4236 individuals in the 2017-2018 National Health and Nutrition Examination Survey dataset were used for model development, and 4748 individuals from the 2015-2016 National Health and Nutrition Examination Survey dataset were used for external validation. We explored the use of decision trees, balanced random forest (BRF), extreme gradient boosting, artificial neural networks, and logistic regression (LR) to distinguish overweight from normal weight. Feature selection was performed with a genetic algorithm, and the data were split into 80% training and 20% validation sets. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration plots, and learning curves, with key metrics including accuracy, precision, specificity, sensitivity, and F1 score. BRF and LR models showed reliable accuracy, generalizability, and stability. The BRF model achieved an AUC of 0.757 on the test set and 0.768 on the training set, while the LR model achieved an AUC of 0.779 on the test set and 0.783 on the training set. External validation showed similar results, with test set AUCs of 0.765 and 0.806 for BRF and LR, respectively. Calibration curves indicated good model calibration and learning curves suggested no significant overfitting. In conclusion, we developed accurate and generalizable machine learning models. Our findings highlight the influence of hepatic and renal function markers and socioeconomic factors on BMI, with low serum phosphorus levels notably associated with high BMI. These results enhance the understanding of serum phosphorus's impact on obesity.