Comparative Analysis of Selected Text Mining Classifiers

Abstract

Text classification is a method of knowledge engineering in which expert level knowledge on classifying documents such that similar documents can be arranged in their respective categories. The field has been a growing area of research as there has been an increase in the availability of textual information. In this paper, three classifiers k-Nearest Neighbour (KNN), Multinomial Naïve Bayes (MNB), and Support Vector Machine (SVM) algorithms were compared based on accuracy and time taken to build a model using five datasets. Experimental results showed that with an increment in the number of training documents, the classification accuracy of the algorithms also increased and algorithms performed better in binary classification tasks than in multiclass classification tasks. The best accuracy of 98.3495% was recorded in SVM using SMS Spam collection dataset compared with other classifiers and datasets, for binary classification, the best classification accuracy of 98.3495% was obtained in SVM using SMS Spam collection dataset, and the best accuracy of 88.28% was obtained in SVM using Reuters 50-50 dataset. The lowest time of 0.01s which was also considered as the best time taken to build a model was recorded in MNB classifier.