Dimensionality reduction for web page classification

Web page classification is the mechanism of assigning a Web page to one or more predefined category. Like other classification tasks, it involves data preprocessing, feature selection, classification and evaluation steps. In the classification step to improve run time performance and accuracy, the classifier must be learned by using representative features selected from the Web pages. The aim of this thesis is to develop some filter-based feature selection methods to reduce dimensionality of Web pages classification problem in order to improve run time performance and accuracy of the classifiers. As supervised classification algorithms, naïve Bayes multinomial, support vector machine and decision trees are used. For the experiments and evaluations Conference and WebKB datasets are used. To evaluate the classification performances of our developed methods, they are compared with classification performances of the well-known filter-based methods that are chi-square, relief, gain ratio, and correlation filter. As performance measure, Macro-average F-measure is used and four folds cross validation is applied. The experimental results showed that, our developed methods are good and can be used as feature selectors for Web page classification.