Edge-Deployable Deep Learning Framework for Intrusion Detection in IoT Networks: A Multi-Model Evaluation Using NSL-KDD, BoT-IoT, and TON_IoT

The Internet of Things (IoT) is a new powerful technological phenomenon with billions of devices implemented in smart homes, healthcare, industry, transport, and urban infrastructure. This dynamic proliferation also grows the network attack surface and sets up security challenges never seen before. The traditional signature-based Intrusion Detection Systems (IDS) are insufficient in dealing with the new, emerging, and zero-day threats in the heterogeneous IoT settings. The proposed paper suggests an intelligent IDS of the IoT networks using sophisticated algorithms of machine learning (ML) and deep learning (DL), i.e., a CNNLSTM hybrid, self-attention BiLSTM (BiLSTM), Random Forest, Support Vector Machine (SVM), and XGBoost, and tested on three benchmark datasets, i.e., NSL-KDD, BoT-IoT, and TONIoT. To tackle the issue of the imbalance of the classes in it and the high dimensionality, the system integrates the SMOTE-based class balancing, the dimensionality reduction through PCA, and the mutual information-based feature selection.