Spatial concentration of objects as a factor in locally weighted models

Abstract

A new approach to construct of spatial econometric models is proposed. It involves the partitioning of objects into groups based on the spatial concentration by k-means clustering. The developed algorithm was compared with known algorithms of k-nearest neighbors and kernel smoothing with a rectangular weight function (kernel). Its significant advantage in running time was shown. The obtained results of computational experiments revealed that the prediction accuracy using the new algorithm yields k-nearest neighbors algorithm but it is about the same as kernel smoothing.