LAPSE: A Machine Learning Message Forwarding Approach based on Node Centrality Estimation in Sparse Dynamic Networks

In graph theory, centrality metrics give information about the importance of a node. Applied to ad-hoc networks, as in the context of sparse dynamic networks, centrality metrics are widely used to take optimal decisions regarding information dissemination. Indeed, nodes showing higher centrality metrics are expected to act as preferred next-hop forwarders in case of dynamic networks. However, in sparse networks, as a consequence of this dynamicity, nodes could choose low centrality forwarder nodes when disseminating their messages, thus resulting in sub-optimal message dissemination.In this paper, we address this problem presenting a Machine Learning-based approach to help estimating the future values of the node's centrality by defining two models based on (i) linear and (ii) polynomial regressions. The node centrality estimation is then exploited in a message forwarding technique we propose called "Linear And Polynomial regression based" (LAPSE). By means of simulations and through the use of real-mobility traces, we show that selecting forwarding nodes via the estimated centrality values allows to obtain better performances than traditional approaches based in terms of the overall centrality of the selected node.