Ensemble methods for text classification

Traditional text classification methods use only one classifier for each text to be classified. Multiple Classifier Systems is a popular research area used to improve classification accuracy. Dynamic Classifier Selection and Dynamic Ensemble Selection are two forms of multiple classifier systems. In Dynamic Ensemble Selection, a classifier ensemble is selected from the classifier pool and their decisions are combined, whereas in Dynamic Classifier Selection, only one classifier is selected from the classifier pool to classify text. In this thesis, existing Dynamic Ensemble Selection and Dynamic Classifier Selection methods have been applied to the text classification problem and it has been shown to increase the classification accuracy in text classification. The other contribution is, a DCS-DQ method, which is a new Dynamic Classifier Selection-based method for text classification, is proposed. In experimental studies, text datasets with different properties were used. The proposed DCS-DQ method is compared with 7 popular Dynamic Classifier Selection methods. According to the experimental results, the proposed DCS-DQ method outperforms the other 7 Dynamic Classifier Selection methods in terms of classification accuracy for the majority of feature sizes. As a result, the proposed DCS-DQ method significantly improves the classification accuracy for text classification.