Comparsion of Classical Discriminant Methods with Artificial Neural Network Using Different Algoritm to the Diagnosis of Myocardial Infarction

Discriminant analysis and classification is one of the most practical statistical parts in various scientific scopes. Classical statistics methods have such prior hypothesizes that they should be considered; otherwise using these methods can cause some noticeable errors. For this reason, researchers tend to choose the methods with less limitation, such as nervation, so they dont need to a specific kind of prior hypothesis. For conducting this research, is used logistic regression method, quadratic discriminant and artificial neural network method for comparison. In this study the participants are 1000 case-control data, who suffered from Myocardial Infarction. Neural network analysis was done with error least squares criterion, and is used Error Back-Propagation algoritm. Then is compared the function of three different BFGS Conjugate, gradient and Gradient descent algorithm of neural network for diagnosis of the stated illness. For analyzing data, is used SPSS, STATISTICA and R-SAS softwares. Based on quadratic discriminant method, prediction error percent, prediction correct percent, sensitivity, specificity and area under the roc curve are 10.15, 89.85, 0.8888, 0.9083 and 0.922, based on logistic regression method these measurements are 10.88, 89.12, 0.8743, 0.9110 and 0.941. And for artificial neural network method these measurements are 3.97, 96.03, 0.9561, 0.9644 and 0.966. There was a meaningful difference between are a under the Rroc curve for these three methods. Also, between three different algorithms, nervation algorithm BFGS has had the best function. The present findings show that the artificial nervation is more accurate for diagnosising Myocardial Infarction illness compared with logistic regression and quadratic discriminant methods.