Implementation of intrusion detection model for DDoS attacks in Lightweight IoT Networks

Abstract

Protecting IoT networks and infrastructure is one of the top priorities in today’s computing industry because of the unnerving and exponential development in cyberattacks and security breaches in IoT. Lightweight IoT networks had been one of the easiest targets for attackers in botnet formation and distributing malware. The research in the paper identifies IoT networks formed by devices with minimal computing resources, such as less battery life, processing power, memory, and more importantly, minimal security, protecting infrastructure, and defense mechanisms, as being lightweight IoT networks that are easily vulnerable to DDoS attacks and disseminating malware. It is investigated by many researchers that development and progress in intrusion detection systems is the need of an hour to safeguard lightweight IoT networks. The manuscript proposes a lightweight Intrusion detection system with a novel data pre-processing technique while using machine learning and deep learning classifiers. The manuscript introduces various types of classifiers, employed to form lightweight intrusion detection systems well suited for protection against Distributed Denial of Services attacks in IoT networks. The datasets used for the experiments and investigation are BOT-IoT and the Network dataset of TON-IoT by the University of New South Wales Sydney (UNSW) Australia. DDoS attack instances are derived from both datasets. Two different experiments are performed on each dataset i.e.; for binary classifications of attack labels, one experiment for all attacks, and another experiment for the DDoS attack only in both datasets. Attack classes in the BOT-IoT dataset compared with the TON-IoT dataset are highly imbalanced. We have used Synthetic Minority oversampling Technique (Smote()) variants for class balancing in the experiments performed on the BOT-IoT dataset.