DODGE: Congestion control in MANET via dragonfly optimized deep learning model

Abstract

A mobile ad hoc network (MANET) is a collection of mobile devices attached without infrastructure or central management. Network size increases rapidly, resulting in congestion, network delay, data packet loss, and a drop in throughput, resulting in poor energy efficiency. Data should be mitigated based on the prediction of congestion. To resolve the problem of congestion, a novel dragonfly optimized deep learning for congestion elimination (DODGE) technique was proposed, which predicts the congested node effectively. Initially, the Transmission Control Protocol (TCP), and User Datagram Protocol (UDP) packets from the MANET environment were pre-processed and the features were selected using dragonfly optimization (DFO). The features that are selected from the DFO model were provided to the stacked convolutional neural network combined with bidirectional long short-term memory (SCNN-BiLSTM). The deep learning network will predict the congested node and if congestion is found, then the message will be displayed. The DODGE is simulated by using Network simulator2 (NS2) and a comparison is made between proposed DODGE and traditional approaches such as hybrid gravitational fuzzy neural network (HGFNN), quality of service-aware distributed congestion control (QoS-ADCC), and improved priority aware ad hoc on-demand distance vector (IPA-AODV) in terms of packet delivery ratio (PDR), delay (DE), throughput (TP), energy consumption (EC), latency (L), detection rate (DR), and network lifetime (NL). The proposed SCNN-BiLSTM improves the overall accuracy better than 10.05%, 6.59%, and 3.26% bidirectional long short-term memory (BiLSTM), deep neural network (DNN), convolutional neural network (CNN) for predicting the congested node in the shortest time.