An approach to text classification using dimensionality reduction and combination of classifiers

Text classification involves assignment of predetermined categories to textual resources. Applications of text classification include recommendation systems, personalization, help desk automation, content filtering and routing, selective alerting, and text mining. This paper describes an experiment for improving the classification accuracy of a large text corpus by the use of dimensionality reduction and multiple-classifier combination techniques. Three different classifiers have been used namely Naive Bayes, J48 Decision Tree and Decision Table. The results of these classifiers are combined using techniques such as Simple Voting, Weighted Voting and Probability-based Voting. The classification accuracy is further improved by the use of a dimensionality reduction method based on concept indexing. Experiments conducted on the Reuters 21578 dataset indicate that the combination approach provides an improved and scalable method for text classification. Also, it is observed that concept indexing helps with classification accuracy in addition to efficiency and scalability.