Enhanced Intrusion Detection for Internet of Things Using Hybrid Algorithms

Computer Science and Engineering

As a foundational technology in the security net environment, detection of network intrusions has a great deal of attention and use. Network Intrusion Detection (NID) still has difficulties when it comes to installing on devices with limited resources, despite the tremendous efforts of research & advanced technologies. We provide a lightweight detection of intrusion technique based on knowledge of extraction is called Lightweight Neural Network (LNet), which strikes the balance between efficiency and accuracy by concurrently lowering computing costs and model storage. To be more precise, we stack DeepMax blocks to create the LNet after carefully designing the DeepMax blocks to extract compressed representation effectively. Additionally, in order to compensate for the lightweight network's performance decrease, we apply batchwise knowledge of oneself distillation to regularise training consistency. Our suggested Lightweight Neural Network (LNN) and XGboost methodology is shown to be successful on Allflowmeter_Hikari datasets through experiments, outperforming the existing methods with less compute burdens and fewer parameters.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.