Online social networks, a significant part of the Internet, contain complex statistical relationships among millions of social nodes. Link prediction algorithms compute a likelihood of potential future ties between nodes in these networks. Most traditional methods use some form of contextual similarity between nodes to discover relations. Traditional community detection models tend to rely on limited structural information and ignore the impact of central nodes during link prediction. These restrictions impair the efficacy of link-prediction-based decision-making. In this paper, a community-based link prediction technique taking the advantages of both is proposed for dynamic and noisy networks. A new community detection metric is proposed to identify candidate links in a web of nodes. The central node in the network is calculated by this metric and used by the link prediction algorithm. The proposed community detection metric, along with other existing metrics, is employed to learn a model for hybrid link prediction. Experimental results indicate that the proposed central node-based link prediction method achieves better performance in prediction accuracy, prediction error, and computation time than traditional methods.