Method of protecting server infrastructure from distributed denial of service attack using a recurrent neural network

Modern statistics from leading cybersecurity companies (Kaspersky Lab, Cloudflare, Statista) indicate a 63% increase in the number of DDoS attacks in 2023-2024. This article discusses an innovative method for countering these threats based on the use of recurrent neural networks (RNN). The developed solution provides automated detection of attacks in real time with the ability to respond instantly, demonstrating classification accuracy at the level of 99.14% and analysis time of only 70.12 ms on a test sample from the NSL-KDD set. A special feature of the proposed approach is adaptability to new types of threats due to the mechanism of additional training of the model. The system can be integrated into the existing client infrastructure without disrupting the operation of services. The conducted studies of various RNN configurations (including analysis of the influence of the sequence length, the number of neurons and training epochs) made it possible to achieve an optimal balance between accuracy and performance.