Blockchain-Enabled Decisive Red Fox Optimizer-Based Feature Selection With Deep Learning-Based Intrusion Detection System

Abstract

Recently, Intrusion Detection Systems (IDS) using Deep Learning (DL) models become useful in accomplishing network security. The selected features serve as input to the DL model, which is trained on labelled datasets to learn intrinsic patterns distinguishing malicious from normal network behaviour. DL techniques like Convolutional Neural Networks (CNN) are usually utilized. The seamless combination of Public Blockchain (BC) technology into the IDS working procedure safeguards a tamper-resistant and safe record of intrusion detection results, improving reliability and transparency in cybersecurity processes. BC is combined into the IDS to improve safety and data integrity. Every legalized intrusion event or classification result is recorded in a decentralized and immutable ledger. BC ensures the reliability of the recognition outcomes, averts tampering, and presents a clear and safe record of network actions. This study introduces a BC-enabled decisive red fox optimizer-based feature selection using the DL (BDRFFS-DL) technique to identify intrusions effectively. The BDRFFS-DL technique exploits the Feature Selection (FS) approach to pick a relevant subset of features, thereby improving classification accuracy and decreasing the computation complexity. Initially, Z-score standardization is used for normalizing the input traffic data into a consistent format. The BDRFFS-DL approach utilizes the DRF optimizer to select the finest feature subset to improve the classification performance and resolve the high dimensionality issue. Furthermore, the intrusion detection process is carried out by using the Convolutional Sparse Autoencoder (CSAE) model. Moreover, BC ensures the integrity of detection results and provides a secure record of network actions. An extensive study of the BDRFFS-DL approach using the ToN_IoT dataset illustrated its superior performance, achieving an accuracy of 98.91%, outperforming existing models.