THE RISK MODELING OF DIABETES BASED ON PARAMETRIC AND NONPARAMETRIC BINARY LOGISTIC REGRESSION

The parametric binary logistic regression assumes that the logit function is known to be expressed as a linear function in the parameter, while the nonparametric binary logistic regression assumes that the logit function is unknown and can be approximated by the Generalized Additive Model (GAM) or Local Likelihood Logit Estimation (LLLE) method. The GAM method assumes that the logit function is the sum of the nonparametric regression functions of each predictor variable with the known link function. The LLLE method assumes that the logit function is a linear function in the parameter, where the parameters depend on arbitrary fixed points and the likelihood logit function depends on the multivariate kernel weighting. In this study we compared the risk prediction of diabetes based on three approaches, i.e parametric binary logistic regression, nonparametric binary logistic regression using the GAM method, and nonparametric binary logistic regression with the LLLE method. The results of classification accuracy in risk prediction of diabetes using the parametric binary logistic regression approach of 80.2%, the GAM method of 88.89%, and the LLLE method of 100%. So, the best approach model is obtained by nonparametric binary logistic regression with the LLLE method.