Common statistical learning algorithms assume independent data points. Among them, the well-known CART algorithm considers independent and identically distributed observations to construct classification rules. In spatial data sets, because of the existence of spatial correlation between observations, the independence assumption is violated, and then it may not be efficient for analyzing spatial data sets. In this paper, we study the classical Boston housing data set using the spatial CART algorithm which is based on both spatial weights and entropy. To do this, latitude and longitude coordinates of the data points from the GIS system as well some covariates are considered for data analysis.

Keywords