A Unified Learning Strategy for Intrusion Identification in Intelligent Cyber-Physical Systems

Abstract

There is an increasing requirement to identify attacks against Cyber-Physical Systems (CPSs) due to their increasing application in critical infrastructures. Despite perhaps depending on sophisticated machine learning (ML) and deep learning (DL) models, the majority of intrusion detection techniques do not take practical implementation into account. This study uses the Knowledge Discovery and Data Mining Cup 1999 (KDDCup99) dataset to develop a novel hybrid deep learning (DL) model for the reliable detection of intrusions in Smart Cyber-Physical Systems (CPSs). The Synthetic Minority Oversampling Technique (SMOTE) is used to address the problem of class imbalance after the dataset has been cleaned, noise-reduced, fitted with outliers eliminated, and normalized using the Z-score. The main innovation is that it blends the temporal sequence modeling of bidirectional long short-term memory (BiLSTM) with the spatial feature-originating of convolutional neural networks (CNNs). The model's ability to capture both sequential and structural attack patterns is made possible by its synergy. With a score of 99.9% on every metric, including F1-score (F1), recall (REC), accuracy (ACC), and precision (PRE), the model achieves exceptionally good results. When compared to more conventional machine learning models like Support Vector Machine (SVM), K-Means Clustering (K-Means) + Random Forest (RF), and a simple CNN, the superiority of the suggested hybrid model is demonstrated. Because of its exceptional generalization and adaptability, it is most appropriate for real-time intrusion detection in the intricate and dynamic Cyber-Physical System (CPS) environment. The potential advantages of hybrid DL in improving the security and resilience of vital smart systems are discussed in this study.