An Approach to Improve the Performance of Web Proxy Cache Replacement Using Machine Learning Techniques

Caching web objects in the proxy server is a well-known approach to improving the performance of web access. Cache replacement policies play a crucial role in the implementation of web caching. There are several cache replacement policies used by different proxy servers. However they are not efficient enough to handle the web requests and also some policies suffered from cache pollution with unwanted web objects. Machine learning approaches recently gained popularity due to their applicability in scenarios where future trends can be effectively predicted from past activities with certain level of confidence. In this work we apply Machine Learning approaches such as SVM and Decision Tree to enhance the performance of the web proxy server accesses. We trained the classifiers using web proxy logs, therefore to predict the class of objects that would be re-visited in the near future. In our proposed approach the unpopular web objects are classified by machine learning classifiers so therefore that they can be evicted before caching. Our empirical results exhibit a promising average hit ratios from SVM and Decision Tree approaches are 51.12% and 67.34% respectively.