NOVEL METHODS FOR DETECTING ANOMALIES IN INFORMATION COMMUNICATION SYSTEMS

As the complexity of ICS increases, security concerns also grow. Anomalies may indicate failures or cyberattacks, making early detection crucial. Traditional statistical and threshold-based methods are ineffective in dynamic ICS environments. ML- and DL-based methods, including support vector machines, autoencoders, and recurrent neural networks, improve accuracy and adaptability. However, they often require high computational resources. Transfer learning, federated learning, and XAI help overcome these limitations. This study analyzes and compares traditional and emerging approaches.