Delay tolerant networks (DTNs) are wireless mobile networks in which the existence of an end-to-end path from the source to the destination of a message cannot be guaranteed. This makes message delivery as one of the major challenges in DTNs. Recent studies based on real world traces show that nodes in DTNs exhibit mobility properties such as their centrality in the network or regularity patterns. To the best of our knowledge, existing routing algorithms exploit only some of the nodes mobility properties (e.g., only nodes centrality, or only nodes regularity) while excluding the others. We present in this paper the first dynamic routing algorithm in DTNs that exploits the most appropriate mobility property (among which node centrality and regularity) at the specific time and location. Our algorithm dynamically learns nodes mobility properties in order to appropriately select the best route to the destination on a per-node and per-situation basis. Simulations performed on real mobility traces show that our algorithm has a better delivery ratio and a lower overhead than existing state-of-the-art routing algorithms that rely on a single mobility property.