Combining classification trees using MLE

We propose a probability distribution for an equivalence class of classific ation trees (that is, those that ignore the value of the cutpoints but reta in tree structure). This distribution is parameterized by a central tree st ructure representing the true model, and a precision or concentration coeff icient representing the variability around the central tree. We use this di stribution to model an observed set of classification trees exhibiting vari ability in tree structure, We propose the maximum likelihood estimate of th e central tree as the best tree to represent the set. This MLE retains the interpretability of a single tree model and has excellent generalizability, We implement an ascent search for the MLE tree structure using a data set of 13 classification trees that predict the presence or absence of cancer b ased on immune system parameters. Copyright (C) 1999 John Wiley & Sons, Ltd .