Machine Learning-Based Optimized Link State Routing Protocol for D2D Communication in 5G/B5G

Device to Device (D2D) communication in Fifth Generation (5G) and unavoidable in Beyond Fifth Generation (B5G) technology is designed to increase network capacity by offloading backhaul links and base stations traffic and improving the performance of low signal nodes, thereby providing fast and energy-efficient communication. D2D communication has several challenges, out of which network extension and data routing to out-of-coverage nodes is the area of focus for this research. optimized Link State Protocol version 2 (OLSRv2) is a popular Mobile Ad-hoc Network (MANET) and Multi-hop D2D communication routing protocol. However, the provision of the D2D-based OLSRv2 routing protocol presents several issues including energy consumption, routing overhead, and optimum relay selection. Therefore, this paper modifies the OLSRv2 protocol by introducing Node’s Status (NS) according to multiple criteria, namely node’s battery level, mobility speed, node degree, and connection to a base station. The proposed routing protocol employed the supervised machine learning (ML) technique to aggregate the multiple criteria into a single comprehensive metric to minimize routing overhead and energy consumption caused by individually transmitting multiple parameters. To determine the best ML techniques for the proposed protocol, four supervised ML techniques, namely K-Nearest Neighbours (K-NN), Random Forest (RF) Multi-layer Perception (MPL), Gradient Boosting Classifier (GBC) has been used to train the model using training datasets generated from the simulation of D2D network in NS-3. The simulation results show that the RF model performed better than the other three models as it consistently reported 100% accuracy and receiver operating characteristic.