Integrating Noun-Based Feature Ranking and Selection Methods with Arabic Text Associative Classification Approach

Abstract

Feature ranking and selection (FR&S) is an important preprocessing phase for text classification, and it is in most cases produces small valuable sub-feature space among the whole feature space and reduces the classification errors. As the associative classification (AC) approach is an efficient method and its training and testing depend on the way that features ranked and selected, the examining of feature ranking methods is very significant. This paper presents an integration method of Arabic noun extraction with four FR&S methods: term frequency–inverse document frequency (TF-IDF), document frequency, odd ratio, and class discriminating measure (CDM). Association rule technology uses the result of the integrated feature selection to construct an Arabic text associative classifier. In this study, the majority voting and ordered decision list prediction methods are used by AC to assign test document to its category. A set of experiments are conducted on collection of Arabic text documents, and the experimental results show that our AC method works better with extracted nouns and feature selection method than with feature selection method individually. The AC based on CDM and TF-IDF methods outperforms the other methods in terms of AC accuracy. As the results indicate, the proposed method produces satisfactory classification accuracy and it has good selecting effect on the Arabic text associative classifier.