Ensemble Machine Learning for Healthcare Data: A Comparative Analysis of Chronic Kidney Disease and Cardiovascular Risk Prediction with NHANES Data

Healthcare survey data, such as the National Health and Nutrition Examination Survey (NHANES), pose analytical challenges due to class imbalances, missing values, and complex variable relationships. This study evaluates data science techniques for analyzing chronic kidney disease (CKD) and cardiovascular health disparities across demographic groups. Black adults face a higher risk of CKD and end-stage renal disease (ESRD), influenced by genetic, socioeconomic, and healthcare disparities. Using NHANES data (2017–March 2020), we examined blood pressure control among non-Hispanic Black, Hispanic, Mexican American, and non-Hispanic White individuals with hypertension, assessing its correlation with protein excretion as a CKD risk factor. We systematically compare traditional statistical methods (odds ratio analysis, hypothesis testing) with machine learning approaches (ensemble learning, feature engineering). Analyzing 666 participants, we implemented weighted ensembles and multi-meta stacking for outcome prediction. Results show ensemble methods achieved superior performance (AUC = 0.864), with blood pressure history as the strongest predictor across models. Feature importance analysis highlights key demographic and clinical variables, while ensemble models enhance predictive balance compared to individual classifiers. This study provides methodological insights for data scientists working with healthcare survey data, emphasizing the advantages of machine learning in handling complex datasets. The findings suggest ensemble models improve prediction reliability and offer a more nuanced understanding of health disparities compared to traditional statistical approaches.