Feature selection optimization of risk factors for coronary heart disease

Abstract

Cardiovascular disease is a worldwide problem and is the main cause of mortality when coronary heart disease leads to a heart attack. Hence, it is important to evaluate how to prevent this disease considering the symptoms description and physical examinations.This study points out the application and comparison of different performance measures for the classification of heart disease. Firstly, a feedforward neural network was applied to classify heart disease risk, using the well-known Framingham database. Feature selection optimization was performed to identify the most important variables to take into consideration, minimizing the Type II error and maximizing the accuracy. In addition, a multi-objective optimization algorithm was carried out to simultaneously optimize both performance measures. A set of non-dominated solutions representing the trade-offs between objectives were obtained, and gender, age, systolic blood pressure, and glucose level emerged as the principal factors to take into consideration to predict heart disease. The results obtained are promising and show the importance of considering more than one criterion to identify the most important variables.