HSL-DCRNet: Hybrid Sequential-Local Deep CNN-RNN Feature Extractor Network for Efficient Intrusion Detection in IoT Network

Abstract

Protecting Internet of Things (IoT) networks necessitates intrusion detection systems (IDS) capable of accurately identifying both temporal behaviors and structural characteristics of malicious traffic. This paper proposes HSL-DCRNet, a Hybrid Sequential Local CNN RNN Feature Extractor Network, to address this challenge. The model employs a Gated Recurrent Unit (GRU) to learn sequential dependencies in traffic flows and a Convolutional Neural Network (CNN) with Inception blocks to extract multi-scale structural features. Their outputs are fused into a unified latent space, and Maximum Relevance Minimum Redundancy (MRMR) feature selection is applied to enhance discriminative power while reducing redundancy. Classification is performed using Enhanced Dense Layers with Adaptive Learning Rate Optimizer (EDL-ALRO), enabling parameter-specific learning rates and faster convergence. Experiments on the UNSW-NB15 dataset show that HSL-DCRNet achieves 99.82% accuracy, surpassing existing IDS approaches. The results confirm its robustness and scalability for securing IoT environments.

Keywords