A Deep Learning-Based Dual-Branch Feature Fusion Model for Network Intrusion Detection

In order to meet the challenge of intrusion detection caused by the complexity and diversity of current network attacks, A deep learning model based on the fusion of two branch features is proposed. The shared convolution layer to extract basic spatial features is employed, then LSTM and GRU are introduced as parallel branches to model long-term dependence and short-term dynamics respectively. On this basis, the cross attention mechanism is used to realize the adaptive fusion of two branch features and enhance the discrimination ability of features. The six-category classification experiment are carried out on the strictly preprocessed CSE-CIC-IDS2018 data set. The verification results show that the accuracy rate of the proposed model is up to 97.96%, and the false positive rate is only 0.09%, which is significantly better than the traditional machine learning method and the single branch deep learning model. Ablation experiments further verified the effectiveness of the double branch structure and fusion mechanism. This work presents a novel technical solution for achieving high-precision, low false-positive intrusion detection in complex network environments.