Hybrid Multi-Task Deep Learning for Improved IoT Network Intrusion Detection: Exploring Different CNN Structures

The rapid expansion of the Internet of Things (IoT) has led to the need for robust security mechanisms to protect IoT networks and devices against various attacks. In this paper, we propose a novel hybrid intrusion detection solution that harnesses the power of multi-task learning (MTL) to enhance intrusion detection performance. We introduce a MTL structure-based model with optimal loss function and task-specific weight optimization, which effectively detects multi-class intrusion threats. Moreover, to address the challenge of imbalanced data in network traffic, we employ a generative adversarial network (GAN)-based oversampling technique for data pre-processing, generating synthetic samples for minority classes. Additionally, we conduct a comprehensive study of different convolutional neural network (CNN)-based deep learning architectures to identify the optimal shared layers for our MTL model, further enhancing the effectiveness of the intrusion detection system. Experimental results on dataset CICIDS2017 demonstrate that despite many rare attacks lacking sufficient number of samples, the MTL-based methodology can deliver superior classification performance.