SDT-IDS: Spatial Data Transformation for Elevating Intrusion Detection Efficiency in IoT Networks

Network Intrusion Detection Systems (NIDS) are critical to identify security attacks or predict an early intrusion activity, a pivotal function in safeguarding the Internet. Recently, Deep learning (DL) has made significant strides in the realm of intrusion detection. Nevertheless, the practical implementation of high-complexity DL models is impeded by the constrained computing capabilities and storage capacity of the Internet of Things (IoT) devices. This article introduces a novel NIDS approach explicitly tailored for IoT, leveraging a lightweight deep neural network. During the data preprocessing phase, to mitigate the issue of high-dimensional raw traffic features that can increase model complexity, we employ a spatially enriched data conversion mechanism to reduce the data dimensionality effectively. Furthermore, when spatial relationships are explored in the data, we can simplify the learning architecture by utilizing the state-of-the-art Vision Transformer (ViT) Techniques, thereby achieving the goal of substantially reducing model complexity and model size. Our experimental results demonstrate that our proposed method achieves impressive accuracy with 99.55% and 99.57% with two traffic input scales. Moreover, we achieved substantial reductions in learnable parameters by 55.35% and 82.07%, along with a remarkable decrease in Floating Point Operations (FLOPs) by 93.56% and 99.28% compared to existing works. This achievement highlights our ability to balance model complexity, size, and performance, making our proposed method highly suitable for deployment on resource-constrained IoT devices.