A comprehensive review of trust aware and optimization assisted machine learning and deep learning intrusion detection systems for mobile ad hoc networks

Mobile ad hoc networks (MANETs) are highly susceptible to routing and insider attacks due to their decentralized architecture, dynamic topology, and limited resources, which undermine the effectiveness of traditional signature- and rule-based intrusion detection systems (IDSs). This review provides a comprehensive and unified synthesis of trust-aware and optimization -assisted machine learning (ML) and deep learning (DL) intrusion detection frameworks developed for MANET security, covering publications from January 2019 to March 2026. Using a PRISMA-based selection process, 89 peer-reviewed studies were included in the analysis. Unlike prior surveys that address a single methodological stream, this review integrates attack taxonomy, observable network indicators, trust mechanisms, optimization strategies, learning models, datasets, evaluation metrics, and deployment feasibility within a single MANET-focused perspective. The review demonstrates that ML-based IDSs offer computationally lightweight detection using routing and forwarding features; DL models such as CNN, LSTM, and GRU capture complex spatial–temporal attack patterns; trust-aware frameworks strengthen resilience against insider and cooperative attacks through behavioural evidence accumulation; and meta-heuristic optimization methods including PSO, GWO, WOA, and GA improve feature selection, hyperparameter tuning, and energy-conscious decision-making. Hybrid frameworks that combine trust evaluation, optimization -assisted feature selection, and ML or DL classification consistently achieve the most favourable trade-off among detection accuracy, false alarm rate, adaptability, and energy efficiency under similar experimental conditions. Practical deployment challenges, including lightweight model design, dataset realism, validation leakage, communication overhead, and scalability under mobility, are critically examined. The review concludes by identifying open research directions in realistic MANET dataset construction, lightweight deep learning, robust trust management against collusion and on–off attacks, federated learning, blockchain-assisted trust transparency, and real-world testbed validation areas.