Feature selection using association word mining for classification

In this paper, we propose effective feature selection method using association word mining. Documents are represented as association-wordvectors that include a few words instead of single words. The focus in this paper is the association rule in reduction of a high dimensional feature space. The accuracy and recall of document classification depend on the number of words for composing association words, confidence, and support at Apriori algorithm. We show how confidence, support, and the number of words for composing association words at Apriori algorithm are selected efficiently. We have used Naive Bayes classifier on text data using proposed feature-vector document representation. By experiment for categorizing documents, we have proved that feature selection method of association word mining is more efficient than information gain and document frequency.