Application of artificial intelligence and soft computing methods for anomaly detection of intrusion attacks in cyber-physical and industrial systems

The digitisation of cyber-physical systems and industrial control systems has amplified the complexity and severity of cyber threats targeting these systems. In particular, intrusion attacks pose a significant threat to the continuity and security of such systems. This thesis investigates the performance of artificial intelligence and soft computing-based methods for anomaly detection of intrusion attacks in cyber-physical and industrial control systems through a systematic approach. To this end, contemporary and comprehensive datasets, namely CIC-ToN-IoT, CSE-CIC-IDS2018, and Edge-IIoTset, were employed to apply various machine learning, deep learning, and soft computing techniques. The performance of classical machine learning algorithms, including decision trees, support vector machines, K-nearest neighbour, logistic regression, and linear discriminant analysis, was evaluated alongside deep learning approaches such as Convolutional Neural Networks and Long Short-Term Memory networks. Furthermore, soft computing methods, including fuzzy logic, Genetic Algorithms, Adaptive Neuro-Fuzzy Inference Systems, and Deep Q Networks, were used to develop both standalone and hybrid models, with their system performance thoroughly analysed. The findings reveal that classical machine learning models generally achieve high accuracy, while deep learning and soft computing-based methods offer advantages in flexibility and adaptability, particularly for minority classes and complex datasets. This study aims to demonstrate the applicability of these methods in cyber-physical and industrial control systems and to provide a foundation for advanced research in this field.