An Intrusion Detection System Using Optimized Deep Learning for IoT Networks

Abstract

In the rapidly evolving landscape of Internet of Things (IoT) networks, assuring security is paramount. Cyber-attacks on IoT networks are evolving in complexity and scale. Intruders employ diverse tactics, including malware, Denial of Service (DoS) attacks, along with unauthorized access. This research aims to propose an optimized Deep Learning (DL) based Intrusion Detection System (IDS) to bolster security within IoT based networks. In order to emphasize the significance of preprocessing data, a critical step is performed to ensure that data is in a suitable format and quality for effective learning and accurate intrusion detection. The preprocessing module performs data cleaning, one-hot encoding and normalization generating normalized inputs for the DL architecture. Subsequently, an efficient Transfer Learning (TL) based Deep Convolutional Neural Network (DCNN) framework is introduced. This framework, characterized by its multi-layered neural networks, autonomously learns and extracts essential features, allowing it to identify unauthorized access attempts and potential malware attacks in an automated and efficient manner. Finally, the neural network training loss is minimized using a hybrid optimization approach that combines Grey Wolf and Improved Salp Swarm algorithms (GW-ISSA). This hybrid algorithm optimizes hyperparameters, leading to faster convergence and reducing the amount of training data required. The NSLKDD dataset is used for simulation using Python, the obtained outcomes contribute to enhancing both security and resilience of IoT networks in the face of emerging threats and vulnerabilities.