Energy-Efficient and Secure Routing Protocol for Mobile Ad Hoc Networks Using Machine Learning-Based Adaptive Path Selection

Abstract

Mobile Ad Hoc Networks (MANETs) provide infrastructure-independent wireless communication through dynamically formed multi-hop links among mobile nodes. Their decentralized operation makes them suitable for disaster recovery, military communication, emergency services, temporary networks, and remote monitoring. However, frequent topology changes, limited battery resources, unpredictable link quality, congestion, and malicious node participation make reliable and secure routing difficult. Conventional reactive routing schemes such as Ad Hoc On-Demand Distance Vector (AODV) primarily establish routes according to connectivity, route freshness, and hop-related information and do not simultaneously optimize residual energy, trustworthiness, link stability, congestion, and predicted route reliability. This paper proposes an Energy-Efficient Secure Adaptive Machine-Learning Routing protocol (ESAMR-ML) for MANETs. The proposed approach extends on-demand routing by evaluating multiple candidate routes using residual node energy, hop count, link stability, traffic load, and direct trust information. A Random Forest classifier predicts the probability that a candidate path will provide successful communication under observed network conditions. Before machine-learning-assisted selection, a deterministic security filter removes routes containing nodes whose trust values fall below a predefined threshold whenever a trusted alternative exists, thereby preventing the learning model from selecting an apparently efficient but insecure route. A multi-objective route score then combines predicted delivery reliability, residual energy, path stability, trust, congestion, and hop count. Preliminary Monte Carlo simulations with 50, 75, and 100 mobile nodes and 12% malicious-node participation show that ESAMR-ML improves average packet delivery ratio from 76.75% with conventional AODV to 84.10%, increases effective throughput from 101.02 kbps to 109.58 kbps, reduces average end-to-end delay by approximately 5.15%, and lowers modeled packet-level energy expenditure by approximately 6.93%. Exposure to routes containing low-trust nodes is reduced by approximately 38.7%. The findings indicate that lightweight machine learning, when bounded by explicit energy and security constraints, can improve adaptive routing in dynamic MANET environments.