A Comprehensive Review of Machine Learning and Deep Learning Techniques for Intrusion Detection Systems

This review synthesizes research on the implementation, technical challenges, and future research directions in applyingMachine Learning and Deep Learning to Intrusion Detection Systems to address limitations in current cybersecuritydefenses against evolving threats. The review aimed to taxonomize ML and DL techniques in IDS, evaluate technicalchallenges including data imbalance and adversarial attacks, benchmark model performance, identify future research priorities,and compare hybrid and ensemble approaches. A systematic analysis of recent studies employing diverse ML/DLarchitectures across network environments was conducted, focusing on detection accuracy, false positive rates, computationalefficiency, robustness, and scalability. Findings reveal that deep learning models, particularly CNNs, LSTMs,and hybrid ensembles, achieve superior detection accuracy but face persistent false positive challenges largely due todata quality issues. Computational overhead restricts real-time deployment, especially in resource-constrained settings,while adversarial vulnerabilities remain critical obstacles with limited practical defenses. Hybrid and ensemble methodsenhance robustness and adaptability but increase complexity and resource demands. The synthesis underscores gaps inexplainability, standardized benchmarking, and scalable real-time implementation. Integrating adaptive, context-awaremechanisms and improving dataset diversity emerge as pivotal future directions. These insights inform the developmentof resilient, interpretable, and scalable ML/DL-based IDS frameworks capable of addressing dynamic cyber threat landscapes.