AI-Driven Cardiovascular Risk Assessment: An Ensemble Machine Learning Approach for Early Disease Prediction

Abstract

Cardiovascular disease (CVD) is the leading cause of mortality globally and, crucially, low cost, easily obtained clinical information can be used to classify patients at risk for early detection in the primary care setting. This paper introduces an AI-based pipeline for cardiovascular disease (CVD) detection that normalizes patient clinical information and classifies cardiovascular risk through an ensemble machine learning system. It is formalized as two very short closed form algorithms: feature standardization and ensemble risk scoring, and evaluated end-to-end on the public UCI Cleveland Heart Disease dataset. The classifiers trained and tested included logistic regression, k-nearest neighbors, support-vector machine, random forest, and gradient-boosted trees.Five classifiers were trained and compared: logistic regression, k-nearest neighbors, support-vector machine, random forest, and gradient-boosted trees. The ensemble methods performed best with random forest attaining 83.6% test accuracy and 0.912 AUC and gradient boosted trees 82.0% accuracy and 0.884 AUC compared to logistic regression (80.3% accuracy, 0.869 AUC). This was validated using 5-fold cross validation (mean accuracy 80.9% ± 4.9%, mean AUC 0.889). Chest-pain type, thalassemia status and ST depression were the most important features identified by the feature-importance analysis, consistent with known clinical risk factors. The results show that the simple, auditable ensemble pipeline is able to provide clinically plausible, competitive classification of the risk of a cardiovascular disease from low-cost clinical tabular data with no imaging or genomic data required.