An Artificial Intelligence of Things Intrusion Detection Framework for Mitigating Cyber and Ransomware Threats in IoT Networks

Abstract

An Artificial Intelligence of Things (AIoT) is an emerging discipline applicable across several sectors, offering a multitude of advantages. AIoT integrates Artificial Intelligence (AI) with Internet of Things (IoT) technology to establish an intelligent network of networked devices, services, and systems. A notable challenge in AIoT security is the presence of multiple vulnerabilities. They were various methods to exploit vulnerabilities and conduct attacks on IoT devices. All cyber-attacks occur via network connectivity, unless one takes into account cyber-physical attacks. IDS (Intrusion Detection System) handle network traffic via devices within an IoT network. It serves as a protective barrier, capable of identifying threats and safeguarding the network against intrusions and malicious attacks. IDS serves as the essential instrument for addressing network intrusions and a range of attacks within contemporary computer network systems. This study aimed to build an efficient IDS utilizing a ensemble model. It builds and trains an ensemble model for time-series or sequence-based classification. It uses 1D convolution layers for feature extraction, and combining predictions from different models utilizing voting rule classifiers. The model is evaluated using standard classification metrics and visualized using a confusion matrix. The testing is done with 80:20 for two data sets (IoTID20 and CIC-IoT2023 with Ransomware attacks) which include the most important types of cyber attacks. This ensemble model is examined against the traditional deep learning models (DNN, DAENN, GRU, LSTM, CNN, BiLSTM and RNN) in terms of Precision, Recall, Accuracy and Fi-Score for both binary classifications and multiple classifications of attacks. The results showed the proposed ensemble model had outperformed the other models with accuracy and reliability reached to 94.63% and 99.99% for CIC-IoT2023 and IoTID20 datasets respectively. Area Under the Curve (AUC) reached to 0.9459 for CIC-IoT2023 and 0.9993 for IoTID20 which indicate the better performance.