An Improved Feature Selection Algorithm Based on Mutual Information
Dandan Dong · Computer Knowledge and Technology · 2009
Feature selection is extremely important research of automatic categorization, and its purpose is to solve the contradiction between the high dimensional feature space and sparse vector of the document. For the less effective classification results of mutual information feature selection method, an improved mutual information feature selection method, IMI,was presented. This method not only takes into the current frequency of feature in text, but also takes into the case of mutual information value is negative. Low frequency words can be filtered more effective. Experiments of automatic categorization based KNN show that IMI improves the classification accuracy.