Risk assessment of coronary heart disease based on cloud-random forest

Coronary heart disease (CHD) is a major public health problem affecting a nation's economic and social development. Risk assessing CHD in a timely manner helps to stop, reverse, and reduce the spread of many chronic diseases and health hazards. This paper proposes a cloud-random forest (C-RF) model combining cloud model and random forest to assess the risk of CHD. In this model, based on the traditional classification and regression trees (CART), a weight determining algorithm based on the cloud model and decision-making trial and evaluation laboratory is applied to obtain the weights of the evaluation attributes. The attribute weight and the gain value of the smallest Gini coefficient corresponding to the same attribute are weighted and summed. The weighted sum is then used to replace the original gain value. This value rule is used as a new CART node split criterion to construct a new decision tree, thus forming a new random forest, namely, the C-RF. The Framingham dataset of the Kaggle platform is the research sample for the empirical analysis. Comparing the C-RF model with CART, support vector machine (SVM), convolutional neural network (CNN), and random forest (RF) using standard performance evaluation indexes such as accuracy, error rates, ROC curve and AUC value. The result shows that the classification accuracy of the C-RF model is 85%, which is improved by 8, 9, 4 and 3% respectively compared with CART, SVM, CNN and RF. The error rate of the first type is 13.99%, which is 6.99, 7.44, 4.47 and 3.02% lower than CART, SVM, CNN and RF respectively. The AUC value is 0.85, which is also higher than other comparison models. Thus, the C-RF model is more superior on classification performance and classification effect in the risk assessment of CHD.