Deep Neural Network-Assisted Multi-Metric Routing for Stable and Efficient Communication in Mobile Ad Hoc Networks

Mobile Ad Hoc Networks (MANETs) are characterized by their dynamic topology, limited resources, and lack of centralized infrastructure. Traditional routing protocols, often relying on a single metric like hop-count, perform suboptimally in such volatile environments, leading to frequent route breaks, high latency, and packet loss. This paper proposes a novel Deep Neural Network-Assisted Multi-metric Routing (DNN-MMR) framework to enhance routing stability and efficiency. The DNN-MMR intelligently synthesizes multiple critical network parameters—including Node Mobility (using Future Link Expiration Time), Residual Energy, and Interface Queue Length—into a composite routing metric termed the "Route Stability Score (RSS)." A Deep Q-Network (DQN) agent is integrated into the route discovery process of a reactive routing protocol (based on AODV) to learn the optimal routing policy by interacting with the network environment. The DQN agent predicts the RSS for potential next-hop nodes, guiding the protocol to select the most stable and efficient paths. Extensive simulations conducted on the NS-3 simulator demonstrate that the proposed DNN-MMR framework significantly outperforms conventional protocols like AODV and its multimetric variants (e.g., AOMDV), achieving a 28% increase in Packet Delivery Ratio, a 35% reduction in Endto-End Delay, and a 40% decrease in routing overhead, while also improving network lifetime by 25% through balanced energy consumption.