Mining user's navigation structure by filtering impurity nodes for generating relevant predictions

Web Navigation Prediction (WNP) has been popularly used for finding future probable web pages. Obtaining relevant information from a large web is challenging, as its size is growing with every second. Web data may contain irrelevant noise, and thus cleaning such data is essential. In this paper, we emphasize the identification and elimination of noisy information from user-navigated web pages. The pruning of user browsed sessions by removing noisy webpages and their relations can help in the development of high-performing prediction models with fewer prediction errors. To minimize the prediction errors, we have proposed four pruning models namely PM3_ER, PM3_EN, PM3_GI, and PM3_EP which remove noisy web pages and their relations from the model consisting of varied simple and complex navigations. The study reveals that PM3_ER is effective in prediction for websites with complicated structures resulting in complex navigations, and PM3_EN is effective in predicting websites with a simple tree-like structure resulting in simple navigations. PM3_ER has shown an improvement in predictions by up to 3% whereas PM3_EN has attained improvement up to 5.45%.