Review of AI-Enhanced Intrusion Detection Systems for Smart City Infrastructures

Abstract

Critical smart-city services—such as intelligent traffic management, power‐grid control, and public Wi-Fi—must be protected against both known and zero-day intrusions. This review paper surveys adaptive intrusion detection systems (IDS) that leverage machine-learning and deep-learning to identify malicious behavior across network, host, and application layers. We classify AI approaches into traditional classifiers (random forests, support vector machines), deep architectures (convolutional and recurrent neural networks), and explainable AI methods that enhance trust and support forensic analysis. The paper contrasts centralized, distributed, and federated IDS architectures, and evaluates trade-offs between detection accuracy, response latency, and energy consumption. We further address regulatory and privacy implications, highlighting how data governance and compliance shape IDS design. Finally, we explore emerging trends such as reinforcement-learning–driven self-healing networks and AI-guided incident response. This concise review equips stakeholders with insights on selecting and adapting AI-based IDS solutions that meet the performance and operational demands of modern smart-city infrastructures.