Feature selection for text classification using OR+SVM-RFE

Feature selection is the key issue in text classification because there are a large number of attributes. In this paper, we propose a new algorithm OR+SVM-RFE that integrates Odds Radio(OR) with recursive feature elimination based on SVM(SVM-RFE). Odds Radio is first used to roughly and rapidly select a feature subset. Then SVM-RFE is used to delicately select a smaller feature subset. Experiment results show the feature subset selected by OR+SVM-RFE obtains a good classification performance with less features.