NEURAL NETWORK BASED CORONARY ARTERY DISEASE PREDICTION USING FEATURE SELECTION BY RANDOM FOREST AND DOMAIN EXPERT

Abstract

Coronary Artery Disease (CAD) has become very common nowadays and is the leading cause of death across the world. According to WHO data published in May 2017 death caused by CAD in Nepal reached 30,559 which is 18.72 % of total death. Angiography is often regarded as best and more accurate method for CAD diagnosis in hospitals however it is more costly and has many side effects. An attempt is being made here for developing a better modal for the classification of CAD by using the clinical, ECG and laboratorial features of patients. Optimum features were selected with the help of domain expert and random forest (mean decreasing accuracy) method. Neural Network based classification model was developed with the selected features for the classification of Left Anterior Descending artery (LAD), Left Circumflex artery (LCX) and Right Coronary Artery (RCA) in human heart. To get the optimum result from the neural network model with 10- fold cross validation method was adopted. The optimized model reached its maximum accuracy of 91.397 % for LAD, 88.09 % for LCX and 90.36 % for RCA classification. Similar new model with common 22 features as input was developed with the data collected from TUTH, Manmohan Cardiovascular and Transplant Centre IOM. Maximum accuracy of 65 % for LAD, 66.67 % for LCX and 60 % for RCA classification was achieved.