A Performance Comparison of Machine Learning Techniques for Cholesterol Prediction

Hypercholesterolemia is an important in cardiovascular disease and it has high risk, but it remains undetected due to unavailability of modern equipment in laboratory for proper screening tests. About 29 million U.S. adults approximately ≈ 1 4% of the total population suffer from cholesterol issues and this indicate that the high demand of alternative screening testing method. To address these major issue we proposed non-invasive method by using ML models to predict the cholesterol level by using method using the NHANES survey data as an input. The data of patient include demographics (age, sex and race/ethnicity), lifestyle factors (dietary intake, physical activity and smoking), anthropometric measures (BMI, waist circumference, blood pressure) and self-reported health history (diabetes, hypertension). The digital data outcome is defined by clinical thresholds (total cholesterol ≥ 240mg/dL or LDL ≥ 160 mg/dL). The data preprocessing involves inputting missing values, encoding categorical variables and scaling numeric features. In first stage we train five classifiers (logistic regression, random forest, XGBoost, SVM, k-NN) using stratified 5 -fold cross-validation, tuning hyperparameters. The performance was tested based on accuracy, sensitivity, specificity and area under the ROC curve (AUC). The results outline that XGBoost model shows the highest discrimination (AUC/ROC ∼0.85) with high sensitivity, outperforming logistic regression (AUC ∼ 0.68) as compared to others. The main data related to predictors included age, BMI, blood pressure, and diet quality, consistent with known risk factors. These results shows that by using survey data meaningfully stratify cholesterol risk can be reduce. Finally, we outline the public health implications based on a non- invasive ML-based screening tool for testing and improving preventive care.