Abstract: Feature selection is the most important preprocessing step in text classification.The quality of the feature words has a significant impact on the accuracy of the classification results.Using the traditional feature selection method,such as MI,IG,CHI,will still cause the redundanly among the feature words.For this problem,based on the combination of the term frequency-inverse document frequency (TF-IDF) and the maximal relevance minimal redundancy (MRMR),this paper put forward a MRMR based feature words secondary selection method TFIDF_MRMR.The experimental results indicate that this method is able to reduce the redundancy of feature words and improve the accuracy of classification.

Key words: Feature selection,Maximal relevance minimal redundancy (MRMR),Term frequency-inverse document frequency(TF-IDF),Text classification