Comparison Some Nonparametric Estimation Methods in Spatial Statistics

Abstract

In this study, we systematically compare four nonparametric approaches in spatial statistics: Local Polynomial (LP), Radial Basis Function (RBF) and its normalized version, Normalized RBF and Thin Plate Splines (TPS). These methods were tested using data of traffic accidents for the year 2023 in the streets network of 15 Iraqi provinces, to assess their predictive ability in modeling spatial distributions of accidents. Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) are among the main statistical measures used in comparing the methods, in which MSE and RMSE were calculated. We found that the strategies were associated with different levels of smoothing, computational efficiencies and flexibility in representing spatial patterns, with Thin Plate Splines striking a good balance between accuracy and flexibility. These results highlight the need for careful method selection that reflects data properties and spatial analysis objectives in order to improve understanding of spatial phenomena and ensure data-driven decisions through sound statistical model selection.