Intelligent web objects prediction approach in web proxy cache using supervised machine learning and feature selection

Abstract

Web proxy cache is used to enhance the performance of network by keeping popular web objects in cache of proxy server for closer access. Intelligent approaches aim at improving the performance of conventional strategies. Mostly focus was on improving prediction mechanism, to guess the ideal objects that will be revisited in future; cache them and combine the result with the conventional algorithm. This research proposes an improved prediction method using automated method to select the influence features that produce accurate prediction results before combining with conventional algorithm. The method use supervised machine learning based on Naïve Bayes (NB) and Decision Tree (C4.5). It applies wrapper feature selection to specify influence features with optimal subset to improve the predictive power. Additionally two more features are extracted to know user's interest to make a smart and a wise decision for caching. The results showed that reduction for the number of features has a good impact on reducing computation time. Moreover, optimal subset selection achieves high performance and enhances accuracy.