Intrusion detection system for IoT networks for detection of DDoS attacks

Abstract

In this thesis, a novel Intrusion Detection System (IDS) based on the hybridization of the Deep Learning (DL) technique and the Multi-objective Optimization method for the detection of Distributed Denial of Service (DDoS) attacks in Internet of Things (IoT) networks is proposed. IoT networks consist of different devices with unique hardware and software configurations communicating over different communication protocols, which produce huge multidimensional data that make IoT networks susceptible to cyber-attacks. The network IDS is a vital tool for protecting networks against threats and malicious attacks. Existing systems face significant challenges due to the continuous emergence of new and more sophisticated cyber threats that are not recognized by them, and therefore advanced IDS is required. This thesis focusses especially on the DDoS attack that is one of the cyber-attacks that has affected many IoT networks in recent times and had resulted in substantial devastating losses. A thorough literature review is conducted on DDoS attacks in the context of IoT networks, IDSs available especially for the IoT networks and the scope and applicability of DL methodology for the detection of cyber-attacks. This thesis includes three main contributions for 1) developing a feature selection algorithm for an IoT network fulfilling six important objectives, 2) designing four DL models for the detection of DDoS attacks and 3) proposing a novel IDS for IoT networks. In the proposed work, for developing advanced IDS, a Jumping Gene adapted NSGA-II multi-objective optimization algorithm for reducing the dimensionality of massive IoT data and Deep Learning model consisting of a Convolutional Neural Network (CNN) combined with Long Short-Term Memory (LSTM) for classification are employed. The experimentation is conducted using a High-Performance Computer (HPC) on the latest CISIDS2017 datasets for DDoS attacks and achieved an accuracy of 99.03 % with a 5-fold reduction in training time. The proposed method is compared with machine learning (ML) algorithms and other state-of-the-art methods, which confirms that the proposed method outperforms other approaches.