Smart Memory Augmented Neural Network for Anomaly-Based Intrusion Detection System in IoT
The Internet of Things (IoT) has enormously developed and is utilized in diverse applications such as healthcare, transportation, military, and agriculture. However, this increasing trend and proliferation of smart objects also make them highly susceptible to malicious attacks. Thus, an anomaly-based Intrusion Detection System (IDS) is employed to prevent attacks by classifying network behavior as normal or anomalous. However, existing IDSs fail to preserve long-term dependencies, and redundant features in the network traffic led to miscalculation. To address these issues, a Smart Memory Augmented Neural Network (SMANN) is developed to observe and remember long-term dependencies during detection by incorporating a memory augmentation framework into the Long Short-Term Memory (LSTM). Furthermore, feature selection is performed using the proposed Fast Flying Particle Swarm Optimization (FFPSO) for selecting highly relevant features by avoiding the problem of oscillation. The separation between the anomalous and non-anomalous patterns of data is ensured by using the fast AnoGAN (f-AnoGAN). To confirm the efficiency of FFPSO-SMANN, the UGR-16, UNSW, NSL-KDD, and CICIDS 2018 dataset are used for assessing and classification analysis. The FFPSO-SMANN is analyzed based on accuracy, precision, recall, and F1-score. The accuracy of FFPSO-SMANN is 99.95% on the NSL-KDD dataset, which is superior to existing methods.