DEVELOPMENT OF MODULAR NEURAL NETWORKS FOR DETECTING DIFFERENT CLASSES OF NETWORK ATTACKS

The article discusses the development of modular neural networks for detecting different classes of network attacks, which is an important step towards improving intrusion detection systems. Modern attack detection systems face numerous limitations, including low efficiency when analyzing large volumes of data, high training time requirements, and challenges in adapting to new types of threats. These shortcomings are due to the use of monolithic approaches, where all network interaction parameters are processed within a single neural network, significantly reducing the system's flexibility and effectiveness. The modular approach proposed in the article involves using separate neural networks to process groups of similar network interaction parameters, which increases attack detection efficiency, reduces model training time, and enables dynamic disabling or retraining of individual modules without stopping the entire system. This architecture allows for more effective attack classification and enhances the system’s ability to adapt to new threats. The article also thoroughly analyzes the advantages of the modular approach compared to traditional monolithic systems, providing significantly greater flexibility and accuracy in detecting and classifying various types of attacks.