A Transfer Learning based Intrusion detection system for Internet of Things
Monika Vishwakarma, Nishtha Kesswani · Research Square · 2023
Abstract There will be billions of gadgets emerging in the future. A few years ago, experts predicted that the Internet of Things (IoT) might soon be renamed the Internet of Everything (IoE) due to the widespread use of computing technologies in modern days. However, what happens if security issues are not addressed in today's IoT devices? Because of cyber security breaches, consumers and manufacturers of connected devices are at risk. Consequently, the number of cyber-attacks has skyrocketed across the networks. Machine learning-based techniques, particularly deep learning, have shown considerable promise in attack detection techniques. This article proposes a 1D Convolution Neural Network (CNN) based model to address anomaly detection in IoT environment. We looked at the capabilities of CNN to identify and categorize abnormalities in IoT networks. The ability of CNN to identify and categorize abnormalities in IoT networks using multiclass and binary classification via transfer learning was also assessed. The performance of the 1D CNN model is assessed using the Netflow-based NF-ToNIoT, NF-BoTIoT, NF-CSE-CICIDS2018, NF-UNSW-NB15, NF-UQ-NIDS, and CIC flowmeter-based IoT DS2, IoT Network Intrusion, MQTT-IoTIDS2020, CIC-ToNIoT datasets. The reason for selecting transfer learning is to reduce classification and run-time complexity. The training and testing times needed for classification are significantly decreased using the transfer learning approach. The proposed model successfully identifies 20 different attacks with an accuracy of 93.75\% on the NF-UQ NIDS dataset. Additionally, we have verified our proposed model in real-time on an edge device with limited resources.