A new deep-learning with swarm based feature selection for intelligent intrusion detection for the Internet of things

Abstract

Humanity can benefit from the Internet of Things (IoT) paradigm in many different ways. IoT devices are nonetheless susceptible to different cyber-attacks launched by the attacker owing to their constrained resources. Using MLTs (Machine Learning Techniques) in IDSs (Intrusion Detection Systems) network data is classified as either benign or harmful depending on its characteristics. However, the growth in IoT devices and the massive, high-dimensional data they produce need the development of more effective searches and learning algorithms. By implementing optimization algorithms on the BoT-IoT dataset, an intruder recognition method for IoT based on DLTs (Deep Learning Techniques). An advanced DLT namely FR-CNN (Faster Recurrent Convolution Neural Network) is used in this framework to categorize incursions. In this paper, a novel GA-FR-CNN (Genetic Algorithm and Faster Recurrent Convolution Neural Network) has been proposed. Additionally, this paper introduces a brand-new feature selection method for IDSs. Network dataset's complexities are greatly reduced and the usage of AAFSO (Assimilated Artificial Fish Swarm Optimization) method to improve recommended systems assisted in identifying characteristics that were important to the problem. After gathering the features, they were sent into GA-FR-CNN algorithms for processing. The experimental result shows the suggested method for an intruder detecting systems, which is based on DLTs, while using UNSW-NB 15 dataset, the AAFSO with GA-FR-CNN provides high accuracy of 94.4880% when compared to the existing methods. When using BOT-IoT dataset, the AAFSO with GA-FR-CNN provides high accuracy of 93.7756%. The proposed method achieves better performance in UNSW-NB 15 dataset than BOT-IoT dataset.