A Two-Stage Predictive Splitting Algorithm in Binary Segmentation

In the framework of binary segmentation, we propose a two-stage splitting algorithm which optimizes a defined predictability function. The idea is to find a binary tree whose nodes are internally most homogeneous and externally most heterogeneous with respect to the predictability of their cases. The main steps of the algorithm will be described. Some relations with the CART splitting procedure will be discussed and an example will be shown.