ADVANCED ATTACK MITIGATION IN IOT GATEWAY PROTOCOLS

Abstract

With the increasing number of users on the internet, numerous cyberattacks are becoming more and more common. Proper detection of these attacks by Intrusion Detection Systems (IDS) is extremely important, particularly for IoT networks. Deep learning methods have proved to be very promising for enhancing IDS performance. This paper presents an end-to-end system for attack detection and prevention in IoT networks with the use of data augmentation, preprocessing, feature extraction, and deep machine learning algorithms. The class imbalance is resolved using the Enhanced Synthetic Minority Over-Sampling Technique (ESMOTE), and preprocessing operations normalize and clean the data for improved model performance. Feature extraction involves statistical features and Shannon entropy-based features, which are fused and sent through a feature selection process. A new 2D-LICM hyper-chaotic map combined with Walrus Optimization (2D-LICMHy-CM_WO) is used to enhance feature selection through enhanced search diversity, convergence rate, and eliminating redundancy. The Dense Convolutional Spatial Attention-based Enhanced Bi-GRU (DCSAtten_EBi-GRU) effectively extracts attack pattern dependencies for precise detection, and an Enhanced Double Deep Q-Learning Network (DoubleDQN) offers dynamic adaptive real-time countermeasures. Experimental findings prove that the proposed solution can obtain a 99.6% detection accuracy with an F1-score of 0.98 and outperforms current IDS models in false positive rate and detection time.