Adaptive Classification of VPN and Non-VPN IoT Traffic Data Using CNN with Spatio-temporal Features

Low network throughput and high round-trip latency are two common challenges in Internet of Things (IoT) communications.This study addresses these issues by proposing an adaptive classification method that maintains network throughput above 360 Mbps and round-trip latency below 1 ms, while achieving a classification accuracy of 90% and a recall rate above 93.33%.This demonstrates the method's ability to provide high throughput, low latency, and efficient traffic classification, addressing critical performance challenges in IoT environments and offering robust support for traffic management and security in real-world IoT deployments.The proposed method utilizes a convolutional neural network (CNN) to differentiate between virtual private network (VPN) and non-VPN traffic.First, the communication traffic data are cleaned using an amplitude-limiting filtering technique to remove noise.The cleaned data are then divided into subsets for preprocessing.Next, a deep learning model is trained using the ISCX VPN-nonVPN public dataset, which integrates both spatial and temporal features of traffic data.Finally, the model leverages the parameter-sharing capability of neurons in the CNN's hidden layers, reducing the number of trained parameters and significantly improving classification performance.Considering the development of new technologies such as VPN intelligent deployment and IPsec policy automatic generation, the research results of this study will accelerate the automation and intelligence transformation of the IoT ecosystem from deployment to management.