Prediction model for coronary artery disease using neural networks and feature selection based on classification and regression tree

Risk of implementing invasive diagnostic procedures for coronary artery disease (CAD) such as angiography is considerable. On the other hand، Successful experience has been achieved in medical data mining approaches. Therefore this study has been done to produce a model based on data mining techniques of neural networks that can predict coronary artery disease. In this descriptive- analytical study، the data set includes nine risk factors of 13228 participants who were undergone angiography at Tehran Heart Center. (4059 participants were not suffering from CAD but 9169 were suffering from CAD). Producing model for predicting coronary artery disease was done based on multilayer perceptron neural networks and variable selection based on classification and regression tree (CART) using of Statistica software. For comparison and selection of best model، the ROC curve analysis was used. After seven-time modeling and comparing the generated models، the final model consists of all existing risk factors obtained with the area under ROC curve of 0. 754، accuracy of 74. 19%، sensitivity of 92. 41% and specificity of 33. 25%. Also، variable selection results in producing a model consists of four risk factors with area under ROC curve of 0. 737، accuracy of 74. 19%، sensitivity of 93. 34% and specificity of 31. 17% was produced. The obtained model is produced based on neural networks. The model is able to identify both high risk patients and acceptable number of healthy subjects. Also، utilizing the feature selection in this study ends up in production of a model which consists of only four risk factors as: age، sex، diabetes and high blood pressure.