Web mining is about combining information collected from the World Wide Web using data mining methods and technologies. Predicting the subsequent web page that may be visited by a particular user has become the most wanted area of research as the need for maximum accuracy is mandatory in the sustenance of business in the World Wide Web. Several models are in use nowadays for prediction which focus on the detecting the users subsequent visit of a web page. Basically all prediction mechanisms concentrate on the basic web usage mining principles of clustering. Obviously we need the services of a prediction model like Markov model which has to be trained with the clusters created by cluster algorithms. Server logs help us to understand the user behavior and the possibility of their next web page visit. In this paper, we prove from the experiments conducted that good formation of clusters will lead to better predictions. Here we present an integrated cluster approach from data sets for the prediction of user requests. This clustering approach defines clusters which can be used for predicting the next user request. We focus on predicting the next request of web users by utilizing Basic Agglomerative Hierarchical Clustering Techniqueor “Bottom-Up” Algorithm and K-Means algorithm along with the prediction algorithm of Markov model. Experimental results reveal that Markov-based models when combined with K-means and Agglomerative approach for clustering produce more accurate results.