A Filter-Wrapper based Feature Selection for Optimized Website Quality Prediction

A Website quality model essentially consists of a set of criteria used to determine if a website reaches certain levels of fineness. Quality attributes are imperative in predicting quality of a website. However large set features may increase model complexity and computational time for prediction. In this study an initial set of 13 quality attributes is prepared and two-stage feature selection based on hybrid filter-wrapper approach is used to optimize website quality prediction performance. In the first stage, each feature is ranked using the Information Gain (IG) method. In the second stage, Firefly Search Algorithm (FA) is applied to features which are ranked higher in the first stage and a feature sub-selection is carried out. In order to evaluate the effectiveness of feature selection method experiments are conducted using four baseline classification algorithms on a dataset prepared by 700 websites. The experimental results show that the proposed model is able to achieve high classification performance. An average of 36.53% reduction in features has been observed which leads to 0.015 seconds of average reduction in time for building a classifier and an average improvement of 5.420% in classification accuracy.