A Hybrid Web Access Prediction Algorithm Using Agglomerative Clustering, Modified Markov Model and Association Rule

The explosive growth of data in the internet makes the people with difficulty in accessing interested pages. Although several methods including Markov model and association rule are available for web access prediction, they have their own limitations in terms of predicting ability and state space complexity. In this paper, it is proposed to identify browsing pattern of people having similar interest using agglomerative clustering approach using k nearest neighbors, modified Markov model and association rule mining. The goal of this paper is to improve prediction accuracy. The homogeneity of clusters is improved very well by exact agglomeration. While doing agglomerative clustering there exist a trade-off between speed and accuracy. The slowness is overcome by reducing the object considered during agglomeration to k, instead of N-1 and by eliminating distant neighbors having similarity value above predefined threshold. Unlike rough sets, this approach considers objects that definitely belonging to a cluster during agglomeration. Hence, cluster validity is improved and computational complexity is reduced. Then, a dynamic Markov model is applied to generate matching states dynamically using cluster for test session. When ambiguity arises, Association rule mining and time-stamp parameter are used to resolve prediction conflicts. The comparative results are presented depicting the improvement in predictive accuracy of the proposed hybrid approach over other systems.