Application of Support Vector Machine for Prediction of Type 2 Diabetes

LI Juan,WU Jiang,LU Li,LIU Dong-lei,PANG Xing-huo,HU Yong-hua

2012-01-01

Abstract:Objective To study the application prospect of support vector machine(SVM) on the prediction of type 2 diabetes(T2D) with environmental and genetic factors.Methods The data from Chinese Twin Registry System in 2001-2004 was used to establish prediction model.Based on the forecasting model of SVM with 18 influencing factors as predictable variables,prediction of T2D was conducted using Matlab software.Results Training accuracy and predictive accuracy via SVM with linear kernel function were 82.50% and 87.50% in prediction model of type 2 diabetes with environmental factors.After considering genetic factors,those accuracy were increased by 1.67% and 2.88%,respectively.The model via SVM with radial basis function was over-fitting with training accuracy as 100.00%,and predictive accuracy as 86.54%.The model via SVM with sigmoid kernel function was inferior to those with linear kernel function,with training accuracy as 81.67%,and predictive accuracy as 86.54%.Considering environmental and genetic factors into the prediction model via SVM with linear kernel function,the sensitivity of 93.33% and specificity of 71.42% were superior to those of model with radial basis function and sigmoid kernel function.Conclusions The model based on SVM with linear kernel function had better effect on predicting the occurrence of T2D.Predictive accuracy of SVM model with environmental and genetic factors was higher than that of model with environmental factors only.SVM had a promising prospect for solving the small sample size and the identification of nonlinear and high-dimension model.