Feature ranking and feature selection algorithms may roughly be divided into three types. The first type encompasses algorithms that are built into adaptive systems for data analysis, for example selection that is a part of neural training. Algorithms of the second type are wrapped around adaptive systems used to determine feature relevance on some data (wrapper approaches). Third type includes feature selection algorithms that are independent of any adaptive system that will be used for classification or approximation, filtering out features that have no chance to be useful in analysis of data. Filter methods are based on performance evaluation metric calculated directly from the data, without direct reference to the results of any data analysis systems. Such algorithms are usually computationally less expensive than those from the first or the second group.

This chapter is devoted to filter methods. It is a draft for the tutorial section of the book "Feature extraction, foundations and applications", edited by Isabelle Guyon, Steve Gunn, Masoud Nikravesh, and Lofti Zadeh, Springer 2006, pp. 89-118, based on the proceedings of the NIPS 2003 workshop on feature extraction and the results of the NIPS 2003 feature selection challenge.

Preprint for comments in PDF, 264 KB.

BACK to the publications of W. Duch.
BACK to the on-line publications of the Department of Informatics, NCU.