5399

5059

كارشناسي ارشد

آمار اقتصادي- اجتماعي

اصفهان: دانشگاه صنعتي اصفهان، دانشكده علوم رياضي

1389

[نه]،119ص.: مصور، جدول،نمودار

ص.ع به: فارسي و انگليسي

سروش عليمردان، مازيار پالهنگ

علي رجالي

8/6/89

حميد پزشك، علي زينل همداني

رياضي

ID5059

به فارسي و انگليسي: قابل رويت در نسخه ديجيتالي

Multilogistic Regression by Means of Evolutionary Product Unit Neural Networks Abstract Nowadays classification is one of the important subjects in many studies According to the significant rule of this subjects in industry medicine and other science we want to improve the classification methods to perform it with more accurate results For instance it is very important to diagnose cancer accurately in a patient The issue of diagnosing cancer is a classification problem in which a patient classified in one of the two categories cancer patient and people without cancer A common method for classification is multilogistic regression Although this method is simple and useful but in the case of the existence of nonlinear effects and interactions between the explanatory variables its accuracy decreases This problem can be solved by inserting appropriate nonlinear functions of explanatory variables in the model In this thesis the nonlinear functions defined as product of explanatory variables raise to arbitrary powers Neural networks and evolutionary algorithms are used to specify the number of nonlinear function which enters the model and also assessing the powers related to the explanatory variables Consequently through some experimental study the model produced by neural networks and evolutionary algorithm compared with the multilogistic regression model The result of the statistical tests determine a significant increase in the classification accuracy We show this results by some steel medicine and industrial data

سروش عليمردان، مازيار پالهنگ

علي رجالي

حميد پزشك، علي زينل همداني