TOWARDS A PERSONALIZED ATHEROSCLEROSIS RISK PREDICTION USING MACHINE LEARNING

Abstract

Accurate and early prediction of atherosclerosis and cardiovascular disease (CVD) is essential for effective intervention. While numerous machine learning approaches have been proposed for this task, the majority rely heavily on lab-based (invasive) clinical variables. These lab-dependent methods often involve delayed results and pose accessibility challenges due to their cost and procedural discomfort. In this study, we develop and evaluate a machine learning framework for predicting atherosclerosis risk using both non-laboratory (non-invasive) and laboratory-based clinical indices. We compare the performance of three classification algorithms – Random Forest, Ensemble (Voting) Classifier, and Multilayer Perceptron - across different input configurations. Experimental results demonstrate that the Random Forest classifier achieved an F-Measure of 95%, AUC of more than 98% using only non-lab features, outperforming the use of lab-based features configurations across all models by at least. 5%. These findings highlight the potential of deploying non-invasive, machine learning-based risk assessment tools as point-of-care applications, enabling early prediction of atherosclerosis without the need for laboratory testing.