Title :
Improved Method of Feature Selection Based on Information Gain
Author :
Li Wei-qiang ; Wang Xiao-feng
Author_Institution :
Coll. of Inf. Eng., Shang Hai Maritime Univ., Shang Hai, China
Abstract :
Feature selection is an essential part of text categorization, which can effectively improve classification precision and efficiency. With some drawbacks proposed from traditional IG approach, an optimized approach that takes concentration and distribution into account is proposed for improving IG approach. The experimental results show that the improved IG approach is superior to traditional IG approach in feature selection.
Keywords :
classification; text analysis; classification precision; feature selection; information gain; text categorization; Accuracy; Educational institutions; Entropy; Information entropy; Random variables; Text categorization; Training;
Conference_Titel :
Engineering and Technology (S-CET), 2012 Spring Congress on
Conference_Location :
Xian
Print_ISBN :
978-1-4577-1965-3
DOI :
10.1109/SCET.2012.6342005
