Machine Learning Models for Cardiovascular Disease: A Holistic Review on Early Diagnosis Performance

This paper evaluates the effectiveness of lightweight machine learning models for predicting cardiovascular disease severity, finding high accuracy and potential for practical use, while emphasizing the need for better explainability and integration with wearable technology.

Cardiovascular disease (CVD) poses a significant threat to global health, responsible for approximately one-third of all deaths worldwide. Early diagnosis plays a crucial role in preventing severe complications, yet traditional methods often require expensive tests and equipment. Machine learning offers an innovative and more cost-efficient alternative, through pattern recognition of more lightweight data to accurately predict CVD presence and severity. In this paper, we look past traditional binary approaches to CVD diagnosis and look towards quantitative ordinal labels, where a higher number represents more severity. Using the UC Irvine Heart Disease dataset, we test six lightweight models — Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbors (KNN), Multilayer Perceptron (MLP), Support Vector Machines (SVC), and Decision Trees (DT) — alongside a custom neural network model to evaluate their performance on this multi-class classification task. Each model's accuracy, precision, recall, and F1 score are analyzed to assess robustness in clinical settings. The RF, KNN, and custom models achieved testing accuracies above 86%, with the KNN leading in accuracy, precision, and recall, going beyond the baseline LR model’s accuracy of 52%. Our findings demonstrate that lightweight models are highly effective for CVD prediction, offering potential for future deployment in AI-assisted tasks in hospitals. However, these models lack explainability behind diagnosis, which is essential for real-world deployment. Future directions include exploring wearable technology for continuous CVD monitoring and the inclusion of worldwide data to account for diverse risk factors.

Explore More!

Research Paper

Source Code