In classical statistics, observations of a random variable are commonly assumed to be independent and identically distributed. Most statistical learning techniques such as Classification and Regression Trees (CART) assume independent samples to compute classification rules. But this assumption is often violated in spatial datasets. In this paper, the CART algorithm is adapted to the case of spatially observations by proposing three strategies; the firrst one is the weighting of the data according to their spatial pattern, the second is spatial Entropy used as the splitting criterion and in the third, we combine these two strategies to achieve more accuracy. Findings are evaluated on a classical dataset to highlight its advantages and drawbacks.

Keywords