Enhancing the performance of SVM based document classifier by selecting good class representative using fuzzy membership criteria

摘要

This work is an attempt to enhance the performance of SVM for document classification using the concept of assigning fuzzy membership to documents in a class. We have proposed a novel way of computing the class representative vectors to get better membership value viz. Preprocessed Mean based FSVM (PM-FSVM). PM-FSVM is based on the concept of preprocessing the mean vector before selecting the class representative (CR) vector with the help of uniformity measure. The strength of our work lies in reducing the effect of outliers and assigning higher membership to the documents which are good representative of their respective classes. The proposed models were compared with standard SVM and FSVM. Experimental results show that our work performs better than existing ones both in terms of recall and precision.