An Investigation Into Cardiovascular Diseases Using Machine Learning Classification Models

Author: Nasri Binsaleh, Mick Christensen, Rithsek Ngem, Onintsoa Ramananandroniaina

Executive Summary

The leading cause of death worldwide is cardiovascular (heart) diseases. In this project, the probability of heart failure was modeled using 918 observations of 11 features linked to heart failure. Thus, the dataset was limited in number of observations and features. The amount of information was limited for a disease that affects millions of people annually, and the dataset was reduced by 20% to create a test set. Thus, the reliability of the models may be less than desired to be applied in practice. A logistic regression model performed the best with an accuracy of 0.87 and a kappa of 0.735, closely followed by a support vector machine model and a random forest. Two MARS and AdaBoost models performed the worst with accuracies of 0.85 and kappas of 0.689. Hence, the recommendation would be to go forward with logistic regression models in the future.

For more information please see FinalReport.pdf