ResTN: Residual Transfer Network for Cross Domain Network Intrusion Detection
Wei Kang Huang, Yong Wang, Zhen Wang · 2025
In the ever-changing network threat landscape, cross-domain intrusion detection in particular has become a key strategy for protecting complex networks from sophisticated attacks that exceed the boundaries of the model’s capabilities. This research endeavors to address the paramount challenge of detecting and mitigating intrusions across diverse network environments and protocol stacks, where conventional systems often falter due to their limited scope and inability to generalize across domains. Building upon the foundations of deep learning, particularly Convolutional Neural Networks (CNNs), our study introduces an innovative framework that integrates advanced Transfer Learning methodologies. In this paper, we propose Residual Transfer Network (ResTN) using a novel transfer learning for the network intrusion detection. The model consists of a two-stage parallel CNN, where the first stage learns on the source domain network intrusion detection dataset and then migrates the model to the target dataset at a small cost. We do carry out various CNN experiments on three datasets. the experimental results show that when transfer to a new data distributions, the key metrics of ResTN, including accuracy, precision, and F1-score, have shown varying degrees of improvement on the UNSW NB15, IOT-23, and CIC-IDS2017 datasets;