Nonparametric Spline Truncated Regression with Knot Point Selection Method Generalized Cross Validation and Unbiased Risk

Abstract

Nonparametric regression approaches are used when the shape of the regression curve between the response variable and the predictor variable is assumed to be unknown. Nonparametric excess regression has high flexibility. A frequently used nonparametric regression approach is a truncated spline that has excellent ability to handle data whose behavior is variable at certain sub-intervals. The aim of this study was to obtain the best model of multivariable nonparametric regression with linear and quadratic truncated spline approaches using Generalized Cross Validation (GCV) and Unbiased Risk (UBR) methods and to find out the factors influencing stunting prevalence in Indonesia in 2021. The data used are the prevalence of stunting as a response variable and the predictor variable used by the percentage of infants receiving Exclusive breastfeeding for 6 months, the percentage of households with proper sanitation, the percentage of toddlers receiving Early Childhood Cultivation (IMD), the percentage of the poor population, and the percentage of pregnant womenIt's a risk. Results show that the best linear and quadratic nonparametric spline truncated regression model in modeling the stunting prevalence is linear truncated spline using the GCV method with three knot points. This model has the minimum GCV value of 7.29 with MSE value of 1.82. Factors influencing the incidence of stunting in Indonesia in 2021 include the percentage variable of infants receiving Exclusive breastfeeding for 6 months, the percentage of households with proper sanitation, the percentage of poor people, and the percentage of pregnant women at risk of KEK.