Multivariate Gaussian Mixture-based Prediction Model for Opportunistic Networks

In this paper, soft clustering on network nodes using Multivariate Gaussian Mixture Models (MGMM) is applied to design a machine learning-based routing protocol for Opportunistic Networks (OppNets). The proposed protocol, called Multivariate Gaussian Mixture-based Prediction routing (MGMP), involves sending messages in concentrated bursts to a group of comparable devices detected using a clustering technique. The network features are utilized for training the MGMM-based clustering model to help identify the relay nodes as the best hop for message transmission. The performance of the proposed MGMP protocol is evaluated using the Haggle Infocom-2006 real dataset and compared against two benchmark protocols KNNR and MLPROPH. The considered performance metrics are average latency, delivery probability and, messages dropped. It has been found that when the TTL is increased, the delivery probability increases and then decreases. Indeed, MGMP outperforms MLPROPH and KNNR in terms of delivery probability by 18.10% and 21.30%, respectively.