Clustering for mobile ad-hoc networks (MANETs) has attracted the interest of many researchers as it offers enhanced scalability and performance improvement. The main challenge of the clustering algorithms is the formation of stable clusters despite the topological changes due to the host mobility. In this paper, we present a novel clustering algorithm, which first predicts the future host mobility and then uses this information to build a stable clustering structure over hosts that will probably exhibit low mobility in the future. In this way, long lifetime for the clustering structure is guaranteed, thereby eliminating the need for frequent reclustering. For predicting the future host mobility, we use provably good information theoretic techniques, which allow on-line learning of a reliable probabilistic model for future mobility.