Feature Selection via Analysis of Relevance and Redundancy
Zheng Lian · 北京理工大学学报：英文版 · 2008
Feature selection is an important problem in pattern classification systems. High dimension fisher criterion(HDF) is a good indicator of class separability. However, calculating the high dimension fisher ratio is difficult. A new feature selection method, called fisher-and-correlation(FC), is proposed. The proposed method is combining fisher criterion and correlation criterion based on the analysis of feature relevance and redundancy. The proposed methodology is tested in five different classification applications. The presented results confirm that FC performs as well as HDF does at much lower computational complexity.