The use of likelihood-based threshold in estimating nonparametric regression models through the adaptive Nadaraya-Watson estimator

Abstract

This study develops a nonparametric regression model using the Likelihood-Based Threshold through the Adaptive Nadaraya-Watson Estimator to model rice productivity data in South Sulawesi in 2023. The issue of data variability can be addressed by simultaneously using bandwidth and threshold to improve estimation accuracy, compared to using only bandwidth. This problem is solved by integrating an adaptive threshold, which allows the estimator to adjust to the characteristics of the data. This method considers the distance between data points and the variation, enabling a more responsive estimation of changes in data patterns. This research aims to obtain the best nonparametric regression model to forecast rice productivity data. The best model is determined using the criterion of the minimum Mean Squared Error (MSE). The analysis results show that the optimal values are h=0.92 and δ=0.99, with the smallest MSE value of 0.075, it produces accurate predictions.