Edge Device Malicious Traffic Detection Based On Machine Learning And Edge Computing

In recent years,with the progress and popularization of high bandwidth and low latency communication technologies such as 5G networks,Wi-Fi 6,and so on,the Internet of Things(IoT)related industries have also entered a high-speed development period.Devices in various fields such as industrial control devices,medical sensors,smart cars,and smart homes are gradually accessing the Internet,affecting various industries,greatly improving industrial production efficiency and people’s quality of life.However,the security risks associated with these devices are gradually exposed to the public.Traditional malicious traffic detection is mainly deployed at the gateway of the upstream network and is detected and warned by the device.However,attacks on edge devices mainly occur in small local area networks,near-source attacks,internal network worm propagation,and other scenarios,and the internal behavior and state of these devices cannot be fully captured by related systems.The invasion detection and malicious traffic detection system of the upstream network also lacks real-time performance,and the attack behavior between edge devices cannot be quickly responded to.At the same time,the existing neural network-based traffic detection structure is relatively complex,requiring more memory resources and computing resources,and many edge devices are IoT devices that do not have the processing capabilities required for related processing.Based on the aforementioned limitations of common traffic detection systems,this study proposes an edge device malicious traffic detection system based on machine learning and edge computing.The malicious traffic detection system is deployed on edge devices such as routers,cameras,and smart homes.It utilizes a low-complexity traffic feature extraction method and a lightweight anomaly detection approach.Additionally,it maximizes the advantages of edge devices by incorporating federated learning into the model training process to accelerate training and improve accuracy.Experimental results demonstrate that even on lowpower devices such as low-end routers and Raspberry Pis,this solution achieves low resource consumption and high accuracy.