Link Pattern Prediction in Opportunistic Networks with Kernel Regression

Opportunistic networks (OppNets) have emerged as prospective network architecture due to the popularization of mobile devices and little monetary cost. Routing is the major concern in designing OppNets. The main obstacle for routing protocol design in OppNets is little knowledge of future link patterns, which leads to blind and unpredictable packet forwarding behavior. To achieve better packet delivery rate, OppNets have to retain and deliver multiple copies of a message, which consumes more devices' energy and causes a lackluster OppNets service. To this end, we aim at the prediction of future link patterns to explore mobile connectivity. In this paper, we propose PreKR-the kernel regression based estimation framework for link pattern prediction. We initially extract best features that can represent the network evolution. Then we models historical structural features by kernel regression with the output of link probability. Experimental results show that our method outperforms state-of-the-art prediction methods up to 25%. We also find that both reach ability prediction and high degree nodes prediction reach more than 90% accuracy. In the end, we propose heterogeneous architecture for PreKR deployment and investigate two prospective OppNets applications to show how PreKR improve system performance.