A Filter Based Feature Selection for Imbalanced Text Classification

Abstract

In this work, a text classification method through a filter type feature selection for imbalanced data is addressed. The model initially clusters the documents associated with a class through a hierarchical clustering there by accomplishing a balanced or near balanced class. Later, a filter type feature selection is recommended to choose the most discriminative features for text classification. Subsequently, the documents are stored in the form of interval valued data. For classification purpose, a suitable symbolic classifier is recommended. The experimentation is done with two standard benchmarking datasets viz., Reuters 21578 and TDT2. The experimental results obtained from the proposed model are better in terms of f-measure when compared to the available models.