On the Use of Neural Networks for the Risk Factor Analysis of NIDDM.
- Author:
Hye Sook SUH
1
;
Jin Wook CHOI
;
Hong Kyu LEE
;
Byong Goo MIN
Author Information
1. Department of Biomedical Engineering, College of Medicine, Seoul National University, Korea. apple@snuh.snu.ac.kr
- Publication Type:Original Article
- Keywords:
Neural Network;
Regression model;
Noninsulin-dependent Diabetes mellitus;
Risk factor analysis
- MeSH:
Artificial Intelligence;
Diabetes Mellitus;
Diabetes Mellitus, Type 2*;
Logistic Models;
Neural Networks (Computer);
Risk Factors*
- From:Journal of Korean Society of Medical Informatics
1998;4(2):127-131
- CountryRepublic of Korea
- Language:Korean
-
Abstract:
There were many cases to apply artificial intelligence to medicine. Neural networks are nonparametric pattern recognition techniques that can be used to model complex relationships. In this paper, we present the analysis of the risk factors of the noninsulin-dependent diabetes mellitus using the artificial neural network and the logistic regression model. First, we developed five prediction models using artificial neural networks and a logistic regression model with the data of Yonchon study of diabetes mellitus. Next, we measured each area under the ROC(Receiver-Operating Characteristic) plots for the performance, and results re followings; multilayer perceptron with seventeen variables(MLP17) was 0.7608, multilayer perceptron with seven variables(MLP7) was 0.7664, radial basis function network with seventeen variables(RBF17) was 0.7919, radial basis function network with seven variables(RBF7) was 0.7715 and logistic regression model(REG7) was 0.8343. All of the variables used are seventeen, and seven variables for neural networks(MLP7 and RBF7) were selected by logistic regression model. The order of higher risk variables in the neural networks(slope) did not completely agree with that in the logistic regression model(odds ratio). However, all of the four higher risk variables that were significant in the statistic model(0.05) also had large slopes(0.3) in the neural network model. And our neural network model also display the influence of another variables in development of the noninsulin-dependent diabetes mellitus.
