Comparison of some methods for estimating a nonparametric regression model (with application)

Abstract

This study aims to estimate a nonparametric regression model using three nonparametric methods: Nadaraya-Watson (NW), K-Nearest Neighbor (KNN), and Spline Smoothing (SS). A simulation approach was used to generate data reflecting real measurements from the Euphrates River, where TDS was the dependent variable and Cl, SO₄, and TH were independent variables from 22 monitoring stations across eight provinces. Results based on MSE showed that the SS method outperformed the others in accuracy and efficiency, and was thus adopted in the applied part of the study.