Attribute Selection Using Artificial Neural Networks – a Case Study of Ischemic Heart Disease

Attribute selection also called as feature selection is a preprocessing technique to select a set of features or subset of features from the available large collection of features. An artificial neural network is the simulation of a human brain which learns with experience. Efficiency of a model or a system in terms of cost, time and accuracy will greatly improve if proper features of a system are selected. This proposed method uses Artificial Neural Network for selecting the interesting or important features from the input layer of the network. A Multi Layer Perceptron Neural Network is used for selection of interesting features from a Ischemic heart data base with 712 patients. Initially the number of attributes was 17 and after feature selection the number of attributes was reduced to 12. All combination of features is attempted as Inputs of a Neural Network. When the input features become 12 the predicted accuracy during training is high as 87.36% using 10 – fold cross validation. Further removal of features lowers the accuracy and hence the interesting attributes selected for prediction is concluded to be as 12 for this Ischemic Heart Disease data set.