Application of Double Sensitive Cost Random Forest in Heart Disease Detection

Abstract

Traditional feature selection algorithms simply compute a feature cost vector to make the random process more tendentious, but do not consider the relative relationship between features, and degenerate into ordinary random forest algorithms when feature differentiation is not significant. In view of this, we propose the dual cost-sensitive random forest algorithm. The algorithm introduces two improvements. 1) Introducing sequential analysis in generating feature vectors, giving dynamic weights to different categories in classification. 2) Introducing cost sensitivity in the decision tree generation stage with the goal of minimum average error. After comparing with logistic regression, random forest, support vector machine and other algorithms, the experimental results show that the method has a lower misclassification rate in heart disease detection, which makes the result classification more reliable and more suitable for practical applications.