Tree-like Dimensionality Reduction for Cancer-informatics

Abstract

Dimensionality reduction in machine learning currently has become a very heated research filed. Traditional dimensionality reduction can be separated into two sub-fields of feature selection and feature extraction, but both of them are under local consideration. In this paper, an algorithm based on information theory, mutual information and maximum spanning tree will be proposed, in order to implement dimensionality reduction under global consideration rather than local consideration. Experimental results show it has a good performance, when the proposed algorithm is applied on gene sequences about cancer bioinformatics.