An effective approach to enhance centroid classifier for text categorization

Centroid Classifier has been shown to be a simple and yet effective method for text categorization. However, it is often plagued with model misfit (or inductive bias) incurred by its assumption. To address this issue, a novel Model Adjustment algorithm was proposed. The basic idea is to make use of some criteria to adjust Centroid Classifier model. In this work, the criteria include training-set errors as well as training-set margins. The empirical assessment indicates that proposed method performs slightly better than SVM classifier in prediction accuracy, as well as beats it in running time.