Classification and Regression Tree (CART) Algorithm for the Prediction of Ischemic Heart Disease

Abstract

CART is a classification technique, which generates a binary decision tree, namely every single interval node having only two children To decide which attributes requires splitting and when it has to be done, CART searches all possible splits of all the attributes and chooses the best split based on the impurity measure of the Gini index method. A Retrospective study contains a total of 7304 patients echocardiography records with one dependent variable and twenty-five independent variables. The “diagnosis” attribute was recognized as a predicted attribute with the value of “1” for patients with IHD and value “0” for the patients with no IHD. There are a maximum of five tree depths, as well as a minimum of 40 and 20 cases in the parent node and in the child node respectively, used in the analysis in order to construct the model and the model thus resulted from the process, portrays information based on the 21 of the total number and 11 terminal nodes, 5 tree depths and 9 parameters were include in the final model for disease prediction. Node 0 is the overall probability of diagnosis; it shows the 1113(15%) proposition of the patients have with IHD and 6191(85%) patients without IHD. FS was identified to be the highly influencing factor for IHD and others attributes i.e. EF, ESV, MR, LVID_s, LA, age, LVID_d, and AV _max are key factors in determining patients with heart disease and strongest interaction with the response variable. This result shows that the IHD of a patient are predicted successfully with an acceptable ratio of 94 %. Furthermore, the true negative rate of the resulting model is high and significant rules were extracted from a dataset that makes the application of Decision tree in predicting IHD in healthcare.