An approach to meta feature selection

Many methods, such as mutual information (MI), document frequency (DF), information gain (IG) and χ² statistics (CHI) algorithm, have been discussed and applied to the study of meta feature selection. This paper gives a brief review of the recent approaches on this topic. By summarizing and synthesizing these approaches, we propose a framework of the application of meta feature selections, where the classical algorithm on attribute reduction is used for data preprocessing and the support vector machine (SVM) algorithm is used for the text classification. The proposed framework has advantages in both effectiveness and accuracy, i.e., our approach decreases the dimension of the text feature space and, at the same time, improves the accuracy of text classification. The experimental results confirm this conclusion.