Artificial Intelligence-Based Intrusion Detection Systems in Smart Networks

Abstract

Smart networks—such as IoT-enabled smart homes, cities, and industrial systems—face a growing threat landscape due to increasing device proliferation and heterogeneity. Traditional signature-based intrusion detection systems (IDS) struggle to adapt to evolving and novel attacks. This paper reviews pre-2019 developments in AI-based IDS tailored for smart networks. We explore both anomaly-based and classification-based approaches leveraging machine learning and AI techniques, including neural networks, fuzzy logic, ensemble learning, and active learning. The study outlines a research methodology involving data collection from smart devices and network flows, preprocessing, feature selection, model training (e.g., supervised or unsupervised), and evaluation based on detection rate, false positives, and response time. Key findings indicate that neural network approaches outperform classical methods, while hybrid systems combining AI with specification-based rules enhance detection. Active learning methods incorporating human analysts boost detection efficiency in IoT contexts. A typical workflow is presented from raw data collection through model deployment in resource-constrained environments. Advantages include adaptability, pattern recognition, and reduced manual rule creation; disadvantages involve computational complexity, data imbalance, and resource constraints. Results and discussion highlight high detection rates (e.g., over 95%) in systems like DÏoT and AI². The conclusion underscores AI’s transformative potential while noting limitations such as interpretability and practicality in constrained devices. Future work proposes federated learning, lightweight models, explainable AI, and continuous learning to strengthen IDS in smart networks. This comprehensive guide captures the state of AI-based IDS before 2019 and sets the stage for future advances.