Comparative study of neural networks and k-means classification in web usage mining

There are many models in literature and practice that analyse user behaviour based on user navigation data and use clustering algorithms to characterize their access patterns. The navigation patterns identified are expected to capture the user's interests. In this paper, we model user behaviour as a vector of the time he spends at each URL, and further classify a new user access pattern. The clustering and classification methods of k-means with non-Euclidean similarity measure, artificial neural networks, and artificial neural networks with standardised inputs were implemented and compared. Apart from identifying user behaviour, the model can also be used as a prediction system where we can identify deviational behaviour.