Enhancing Cybersecurity: The Development of a Flexible Deep Learning Model for Enhanced Anomaly Detection

Using expert systems and relevant machine learning methods, automating network intrusion detection has become commonplace. However, the interconnectedness of many industrial control systems and the Internet of Things (IoT) has made cyber attacks on critical infrastructure communication networks a significant concern. These Critical Cyber-Physical Systems (CPSs) experience massive network traffic, posing a challenge for conventional machine-learning techniques to identify anomalies. This paper introduces a novel approach that overcomes these limitations, leveraging the power of deep learning to detect and classify anomalies with remarkable accuracy. By utilizing deep models like Deep Neural Network (DNN) and Deep Short-Term Memory (LSTM) in a two-stage procedure, we significantly enhance the capabilities of our proposed methodology. We also employ a Deep Sparse Autoencoder (DSAE) to resolve the feature engineering problem and prepare data for processing. The effectiveness of this strategy is evaluated using datasets collected from the IoT ecosystem, specifically IoT-23 and LITNET-2020. The evaluation results for the proposed method are presented and discussed, comparing its statistical significance with the most cutting-edge techniques for detecting network anomalies.