A differential evolutionary architecture for artificial neural trees with applications to medical data mining

This paper presents an evolutionary structure for neural trees by differential evolution. For a neural tree a structure tree and weights tree are defined. A partial ordering based multi-dimensional fitness function is applied to measure the energy of the system. Different to traditional evolution method of genetic programming, a new evolution technique based on differential evolution is proposed. A numbering scheme is designed for basic operations of tree structure. Finally we present an experimental framework for the tree evolution. Application frameworks are given in medical data mining and analysis.