Stemming text-based web page classification using machine learning algorithms: a comparison

Abstract

The research aim is to determine the effect of word-stemming in web pages classification using different machine learning classifiers, namely Naive Bayes (NB), k-Nearest Neighbour (k-NN), Support Vector Machine (SVM) and Multilayer Perceptron (MP). Each classifiers' performance is evaluated in term of accuracy and processing time. This research uses BBC dataset that has five predefined categories. The result demonstrates that classifiers' performance is better without word stemming, whereby all classifiers show higher classification accuracy, with the highest accuracy produced by NB and SVM at 97% for F1 score, while NB takes shorter training time than SVM. With word stemming, the effect on training and classification time is negligible, except on Multilayer Perceptron in which word stemming has effectively reduced the training time.