AI-driven solutions for proactive network security and threat detection
Gang Li, Yan Sun, Hai Fu, Yaowen Sun · IET conference proceedings. · 2025
This paper introduces the Adaptive Threat Detection Network (ATDN) designed to address the complex and diverse cyber threats in university networks. ATDN leverages reinforcement learning, multi-modal data fusion, and adversarial training to improve the detection of both known and unknown threats. We evaluated ATDN using three open-source datasets—NSL-KDD, CICIDS2017, and UNSW-NB15—and compared its performance against several baseline algorithms. The results show that ATDN significantly outperforms traditional models in terms of accuracy, precision, recall, and F1-Score. The ablation study highlights the key roles of reinforcement learning and multi-modal data fusion in the model’s performance. ATDN efficiently handles the complex traffic patterns in university networks, maintaining low latency for real-time response. Future research will focus on extending ATDN to handle encrypted traffic and incorporating additional data sources to further strengthen its ability to mitigate modern network threats.