Machine Learning-Driven Congestion Prediction in Mobile Ad-Hoc Networks Through Modelling Approaches

Abstract

Mobile Ad Hoc Networks (MANETs) operate autonomously through decentralized configurations for military as well as emergency and academic applications. The adaptable network structure and unstable nature of MANETs result in major traffic jam occurrences when network activity is at the peak. This research studies the congestion issue of MANETs by implementing network simulation with Machine Learning analytics to identify and control traffic congestion effectively. The investigation employed OPNET 14.5v to simulate office scenarios that contained five, ten and fifteen mobile nodes to study congestion patterns. The study measured network performance through three metrics that consisted of bits per second network load and seconds of media access delay as well as bits per second traffic reception. Network congestion increased as node density increased because 2.8 Mbps load appeared under five nodes but the network load reached 5.2 Mbps with fifteen nodes. Maximum traffic conditions caused media access delay to reach its highest point at 0.0056 seconds. A collection of ML models included Decision Trees and Random Forest followed by Artificial Neural Networks (ANNs) for congestion detection purposes. The evaluation resulted in substantial experimental precision levels of 98.7%, 99.3% and 99.8%. This research proved that using ML-based adaptive load balancing promoted both network stability along with real-time throughput enhancement when faced with congestion situations. The findings prove that predictive analysis operates in real-time to solve traffic congestion problems which results in improved routing stability and decreased delays in military ad hoc networks. Through the OPNET simulation platform researchers gain an organized environment to evaluate and enhance such systems.