Intelligent Intrusion Detection System Based on Hierarchical BiGRU and Advanced Feature Selection Techniques

An Intrusion Detection System (IDS) is a product that can identify the activity across more networks and alert the stakeholders to potentially harmful threats or malicious actions. Thus, the advancement in improving IDS is introduced in developing an Intelligent IDS (IIDS) centred on Deep Learning (DL) techniques for feature extraction. Moreover, the IDS methods generally underwent the issues such as false positive rates, adaptability to change and compute resources that can lead to ineffective detection of potential threats to a network during the real-time conditions. Hence, the proposed model uses a multi-strategy Sparrow Search Algorithms (SSA) and a Variational Mode Decomposition (VMD) that uses feature selection to select only the most important features, which reduces dimensionality when processing the data, by reducing the prior false positive rates accuracy. Moreover, a Hierarchical Bidirectional Gated Recurrent Unit (BiGRU) architecture with residual gates and a Time-Aware Attention Mechanism is used. It allows the model to more effectively capture the complex temporal dependencies and contextual information found in the network traffic data. The method is tested using the CIC IDS 2018 dataset which a benchmark dataset is containing many of the modern attack types so that it can be confident in the impact of the proposed method. The model enhances real-time network security from an advanced feature selection approach and a complex hierarchical BiGRU architecture that is adaptable, scalable and efficient. The advantages of the proposed model are demonstrated by its performance metrics with its high accuracy, precision, recall, and F1-score, signifying that the model is able to effectively identify the different intrusion types while also minimizing false alarms.