Discovery of User Navigation Patterns on a web site through Data Mining Algorithms

Web Mining is the application of data mining algorithms to mine significant patterns from the web. Web usage mining is the technique used to identify the user's need and interest on the World Wide Web. This paper highlights the effect of classification on the server logs from the msnbc dataset for the month of September 1998. This dataset contains information on 65536 users and their navigation behavior through the 17 web pages (each focusing towards frontpage, news, tech, local, opinion, on-air, misc, weather etc,). In order to have efficient classification patterns, the original msnbc data is pre-processed, leading to subsets, each focusing towards news, health, on-air, weather, etc,. Hence the class fields in each of the training set correspond to any one of the core pages (fi-ontpage, news, etc,) resulting in 17 subsets. This research aims at identifying important patterns in the usage of web pages listed in this data and brings out the interest of the users while navigating the web site. We have evaluated the performance of eight classification algorithms on the msnbc dataset and report higher accuracy for the Quinlan's C4.5 algorithm and the Random Tree algorithm. The error-rates revealed by the algorithms indicate the usage density of the core pages (News, Weather, etc,). Moreover the misclassification rates indicate the usage style of different users on a web page with less error being generated for pages with more user hits that unearth the user navigation patterns.