Using supervised Neural Network s For Heart Diseases Diagnosis

Abstract

Pattern recognition is one of the artificial intelligence branches. Neural networks are the most popular method in pattern recognition. Neural network is considered a promising tool for recognizing disease from its symptoms. In this research a heart diseases neural-network-based classification system is designed and trained to classify four types of heart diseases (Healthy, coronary artery disease, Stroke and Valve). The architecture of this network is feed forward trained by back propagation training method. Matlab version 7 tools box is used to design, train and test this artificial neural network. University of California, Irvine (UCI) repository Heart disease data set has been used to test and train the proposed network *. This data set contains 303 data samples, which has 14 symptoms for each sample. The result obtained from training phase is about 99.8% and from testing phase we obtained accuracy around 84%. The experiments indicate that increasing the data set size and/or enhancing the network architecture and training procedure could enhance the result.