PERFORMANCE COMPARISON OF SEMIPARAMETRIC KERNEL REGRESSION USING EPANECHNIKOV, GAUSSIAN, AND, BIWEIGHT KERNELS IN MODELING LIFE EXPECTANCY IN INDONESIA

Life expectancy is a crucial indicator of population health and socioeconomic development, and its relationship with explanatory variables often exhibits both linear and nonlinear patterns. This study employs a semiparametric kernel regression approach to model life expectancy in Indonesia by integrating parametric and nonparametric components. Undernutrition is modeled parametrically, while access to improved drinking water and the illiteracy rate are treated as nonparametric predictors to flexibly capture their nonlinear effects. The nonparametric component is estimated using the Nadaraya–Watson kernel estimator, whereas parameter estimation is conducted using the Ordinary Least Squares method. Several kernel functions are evaluated, and the optimal kernel and bandwidth are selected based on the minimum Generalized Cross-Validation (GCV) criterion. The results indicate that the Gaussian kernel with an optimal bandwidth of (100, 100) yields the best performance. The proposed model demonstrates excellent accuracy, with a coefficient of determination of 99.99% and a very low Root Mean Square Error (RMSE). These findings confirm that semiparametric kernel regression is a flexible and reliable method for modeling life expectancy and other public health indicators characterized by complex relationships.