Optimizing multivariable kernel regression with GCV and UBR for systolic blood pressure modeling in type 2 diabetes mellitus

Abstract

Random blood glucose and age can influence the increase in systolic blood pressure in patients with type 2 diabetes mellitus. The data pattern forms a random pattern, so the modeling uses multivariable kernel nonparametric regression. This study uses the Nadaraya-Watson estimator because the calculation is simpler and uses the Gaussian kernel function with bandwidth optimization methods are Generalized Cross-Validation (GCV) and Unbiased Risk (UBR). The model with GCV optimization obtained an \(R^2\) of 99.9713%, while the model with UBR optimization obtained an \(R^2\) of 99.9726%, so the model produced by UBR optimization was the best. The model with UBR optimization revealed that systolic blood pressure in patients with type 2 diabetes mellitus was strongly influenced by random blood glucose and age. Evaluation of the UBR optimization model performance on the testing data yielded an Symmetric Mean Absolute Percentage Error (SMAPE) value of 18.05096%. This means that the model performance with UBR optimization has a good ability to predict data.