AI 학습기반 네트워크 공격대응 시스템 설계 = Design of AI-Based Network Attack Response System

다국어 초록 (Multilingual Abstract)

Recently, network usage has increased across various sectors, and the level of dependence on network infrastructure throughout industry has grown significantly. However, alongside this expansion, various adverse effects have also risen, among which performance-degrading attacks such as DoS, DDoS, and UDP Flooding have become increasingly severe. In this paper, to address these issues, DoS, DDoS, and UDP Flooding attack scenarios were reproduced in an NS-3-based virtual network environment, and the detection performance of an Intrusion Detection System (IDS) was analyzed. Detection logs were recorded in CSV format and utilized to design a TensorFlow-based reinforcement learning model aimed at improving the true positive rate while reducing the false positive rate. Furthermore, Explainable Artificial Intelligence (XAI) techniques were applied to visualize the detection process, thereby ensuring transparency in the model’s decision-making rationale. In particular, the trained model was integrated with a real-time alert mechanism so that operators would receive immediate notifications upon detection of abnormal patterns.