Prediction of metabolic syndrome in mexicans using machine learning

Abstract

Metabolic syndrome (MetS) is a compelling public health issue in Mexico, with high prevalence rates of overweight, obesity, arterial hypertension, diabetes, high triglycerides, low high-density lipoprotein cholesterol, and high total cholesterol. Despite this, predictive models tailored for under-researched professional groups with sedentary habits are scarce. This study introduces a novel predictive model for MetS using data from the National Center for Health Statistics and a unique dataset of higher education staff. By employing and comparing machine learning algorithms such as decision trees, random forest, artificial neural networks, and adaptive boosting, the research provides new insights into gender and race-specific aspects of MetS. The data was labeled using standards from the International Diabetes Federation and the National Cholesterol Education Program Adult Treatment Panel III to create classification models, which were tested on the higher education staff dataset. Model predictions were assessed using F1-score, accuracy and area under the curve - receiver operating characteristic (AUC-ROC), with random forest, decision tree, and adaptive boosting performing best. The key predictive features identified for MetS prediction include triglycerides, glucose, high-density lipoprotein cholesterol, waist-to-height ratio, and body mass index.