Artificial Intelligence-Driven Computational Models And Biomathematical Algorithms For Predictive Healthcare Analytics, Disease Diagnosis, And Smart Medical Decision Support Systems

Abstract

Cardiovascular disease remains a major public-health burden, requiring scalable and interpretable tools for early risk prediction and smart medical decision support. This study created an integrated framework based on artificial intelligence and biomathematical models to predict the risk of self-reported coronary heart disease (CHD) / myocardial infarction (MI) using the Behavioral Risk Factor Surveillance System (BRFSS) dataset 2024. Following pre-processing and outcome filtering, 452,464 respondents were included, and 9.36% of them had CHD/MI. Stratified train-test splitting, survey weighting, threshold optimization, probability calibration, explainability analysis and assessment of subgroups were used to train and evaluate four supervised models: weighted logistic regression, XGBoost, LightGBM, and CatBoost. CatBoost performed best in the held-out test set in terms of discrimination (AUROC = 0.8392; AUPRC = 0.3506). The CatBoost Brier score was reduced after Platt calibration from 0.1650 to 0.0712 and the final calibrated threshold was 0.08 with a sensitivity of 0.8250, a specificity of 0.6935, a negative predictive value of 0.9746, and a balanced accuracy of 0.7592. The Biomathematical Cardiovascular Risk Index was an interpretable comparator, with an AUROC of 0.7971. Clinically meaningful predictors were identified by SHAP analysis such as age, general health, diabetes, smoking, kidney disease and BMI. In summary, the calibrated CatBoost-BCRI is suitable for cardiovascular risk stratification and preventive decision support with interpretability, but needs further validation prior to clinical use.