kROp: k-Means clustering based routing protocol for opportunistic networks

Routing and forwarding of messages is a challenging task in Opportunistic networks (Oppnets) due to no fixed architecture and the intermittent connectivity of the nodes. This paper proposes a context-aware routing protocol named kROp, which uses a variety of network features generated dynamically for making routing decisions. Four network features have been defined in this work which is vital for identifying potentially good message forwarders. kROp utilizes unsupervised machine learning in the form of an optimized k-Means clustering algorithm to train on these features and make next hop selection decisions. The performance of kROp is evaluated and compared with HBPR, ProWait and PRoPHET routing protocols in terms of hop count, dropped messages, delivery probability, and overhead ratio. Index terms: kROp, Opportunistic Routing, Unsupervised Learning, k -Means Clustering, Context-Aware Routing, Delivery Probability, Average Hop Count, Wireless Networking.