Interval-Valued Feature Selection for Classification of Text Documents.

This paper presents the classification of text data using the symbolic data type of interval-valued feature selection method. Initially, the documents are represented in the form of interval-valued features. The proposed method uses a supervised environment in which every feature is represented using a single crisp value with the help of the proposed ranking method. Further, the features are ranked using scores associated with each of them. The top-ranked Q′ features are chosen from the Q set of evaluated features, and Q′ is decided through empirical evaluation. The feature selection criteria proposed is validated using symbolic classifier with the help of standard text datasets Reuters-21578 and TDT2 dataset. The experimental results obtained from this method show that the proposed method is more effective compare to other existing techniques.