Intrusion Detection Model Based on Recursive Gated Convolution and Bidirectional Gated Recurrent Units

Abstract

To address the issue that convolutional neural networks (CNNs) struggle to capture long-range feature dependencies and deep temporal features, this paper proposes a deep learning model based on recursive gated convolution and bidirectional gated recurrent units (Bi-GRU). The model fully considers the characteristics of network traffic data and designs a dual-path feature extraction mechanism. The first path extracts local features at different scales by stacking standard convolutional kernels of various sizes, while the second path uses recursive gated convolution to capture long-range feature dependencies. The features extracted from both paths are fused through element-wise multiplication. Subsequently, deep temporal features are modeled using Bi-GRU, and the final prediction results are obtained via a Softmax layer. Comparative experiments on the NSL-KDD dataset show that the proposed model performs exceptionally well in the comprehensive evaluation metric Weighted_F1, outperforming other comparative models.