AI-Driven ML and DL Approaches for Internet of Things Security in Developing Countries: A Systematic Review

Abstract

The Internet of Things (IoT) has significantly expanded the attack surface of modern digital infrastructure. In developing regions, low-cost devices are deployed across fragile networks for precision agriculture, public utilities, clinical settings, and city governance, often with minimal cybersecurity oversight. These conditions substantially elevate the risk of cyberattacks. Traditional intrusion detection systems impose computational and memory requirements that microcontroller-grade IoT hardware cannot satisfy. In this study, we systematically surveyed the use of lightweight AI, ML, and DL methods to protect resource-constrained IoT (RC-IoT) devices. Benchmarked solution classes range from shallow classifiers and compressed neural networks to edge-gateway models and federated learning frameworks, which are evaluated based on their accuracy, memory footprint, inference latency and energy consumption. Well-tuned models can maintain high detection rates without exceeding the hardware limitations. A fundamental trade-off exists between detection accuracy and energy consumption: deeper models that improve detection fidelity also increase power demand, a critical concern for battery-operated nodes in settings where reliable grid power cannot be assumed. Techniques such as federated learning and compact cryptographic primitives enable privacy-preserving coordination of distributed nodes. Grounded in a layered defense architecture, this review concludes with actionable guidance for engineers, institutions, and policymakers in resource-constrained environments.