Network Security for Cyber-Physical Systems Using Deep Neural Network-Based Anomaly Detection

Abstract

In recent years, Cyber-Physical Systems (CPS) have seen explosive growth in popularity thanks to their many practical uses. Network security and user privacy are key concerns while deploying CPS networks because of the high number of internet-connected devices in such an ecosystem, which makes them more susceptible to cyber-attacks. An effective and efficient Intrusion Detection System (IDS) might be a feasible way to defend CPS networks from different threats. This study proposes a new intrusion detection system (IDS) for Cyber-Physical Systems networks that uses deep learning to detect anomalies. In particular, we have introduced a Deep Neural Network (DNN) model for filter-based feature selection that drops features with strong correlations.

In addition, several parameters and hyperparameters are used to fine-tune the model. The UNSW-NB15 dataset, which includes four types of attacks, is used for this. To address class imbalance concerns in the dataset, the suggested model was trained using Generative Adversarial Networks (GANs). It then generated synthetic data of minority assaults and attained a 98% accuracy rate with the balanced class dataset.