Institut für Informatik

Technical Report No. 68, April 1995 - Abstract

Alois P. Heinz, Christoph Hense:
Tools for Neural Trees

Ntree is a C library of tools for the construction, training, and evaluation of Neural Trees. It contains an improved version of the CART algorithm for the construction of binary decision trees that incorporate splits with respect to fuzzy sets. These trees can be regarded as (or translated to) Neural Trees. For their training the Ntree library implements the AFLN algorithm. With these algorithms one is able to learn functions of the kind f: D1 x ... x Dn -> R from a given set of training data where for each i in {1 ... n}, Di may either be R or an unordered set of symbolic attribute values. This document mainly describes the usage of the Ntree library.