Open AccessUNBIASEDNESS OF FEATURE SELECTION BY HYBRID FILTERING
Author(s) -
Wiesław Pietruszkiewicz
Publication year - 2011
Publication title -
computing
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.184
H-Index - 11
eISSN - 2312-5381
pISSN - 1727-6209
DOI - 10.47839/ijc.10.1.735
Subject(s) - feature selection , computer science , feature (linguistics) , selection (genetic algorithm) , voting , focus (optics) , machine learning , artificial intelligence , data mining , philosophy , linguistics , physics , optics , politics , political science , law
In this article we examine characteristics of feature selection algorithms by introducing their aspects important in practice. We will focus on the unbiasedness, analyse it and investigate a robust hybrid method of feature selection, being a composition of several feature filters, that could ensure unbiased results of selection. Using parallel multi-measures and voting, we reduce the risk of selecting non-optimal features, a common situation when we select attributes using single evaluation based on one evaluation criterion. To test this method we selected a personal bankruptcy dataset, containing various types of attributes and one of the popular machine learning benchmarks. By the performed experiments we will demonstrate that an approach of multi-evaluation used for features filtering may lead to the creation of effective and fast methods of features selection with an unbiased outcome.