Satin bird optimization and sliced Bi-directional gated recurrent unit based network intrusion detection system

In recent times, the Intrusion Detection System (IDS) have played a crucial role in enhancing security and detecting anomalies in networks. Deep Learning (DL) algorithms have demonstrated high efficiency in capturing optimal features and achieving more accurate differentiation between normal and attack classes. As the amount of data increases, high-dimensional features make the training process more difficult. To overcome this, this article develops the Satin Bird Optimization (SBO) and Sliced Bi-directional Gated Recurrent Unit (SBi-GRU) technique to detect intrusions in a network. The SBO algorithm is designed to select significant features from entire feature set, thereby reducing the feature dimension and enhancing classification performance. The SBi-GRU network includes a slicing process, bi-directional structure, and GRU network. The slicing mechanism accelerates the training process and balances the effectiveness and performance. Subsequently, Multi-Head Self-Attention (MHA) is integrated with SBi-GRU to learn hidden patterns across various subspaces. The developed SBO and SBi-GRU algorithm achieved 99.99% accuracy on NSL-KDD, 99.99% accuracy on CIC-IDS 2018, and 98.98% accuracy on UNSW-NB15 when compared to other conventional algorithms such as Long Short-Term Memory (LSTM) and Auto-Encoder (AE).