Abstract
Billions of IoT devices increasingly function as gateways to cloud infrastructures, making them an inevitable target of cyber threats because of the limited resources and low processing capabilities of IoT devices. This paper proposes a lightweight decision tree-based intrusion detection framework suitable for real-time anomaly detection in a resource-constrained IoT environment. Finally, the model also makes use of a novel leaf-cut feature optimization strategy and tight adaptive cloud edge intelligence to achieve high accuracy while minimizing memory and computation demand. In terms of memory, they also use only 12.5 MB in it and evaluated on benchmark datasets including NSL-KDD and Bot-IoT, it gives an accuracy of 98.2% and 97.9%, respectively, and less than 1% false positives, thereby giving up to 6.8% accuracy over some traditional models such as SVM and Neural Networks and up to 78% less energy. It is deployed on Raspberry Pi nodes and can do real-time inference in less than 1 ms and 1,250 samples/sec. Due to the energy efficient, scalable, and interpretable architecture of the proposed solution, it can be implemented as a security solution for IoT use cases in Smart cities, industrial automation, health care, and autonomous vehicles.