Using CNN to Optimize Traffic Classification for Smart Homes in 5G Era

Abstract

With the rapid development and progress of the Internet of Things and artificial intelligence, more and more businesses have combined housing with emerging technologies to create smart homes to improve residents' quality of life. Many services similar to the three major application scenarios of 5G will be applied to different smart devices in future smart homes. Therefore, the overall network traffic of smart homes will inevitably increase substantially, making network traffic management in smart homes an issue worthy of in-depth discussion. However, due to the widespread use of network encryption, it is not easy to obtain information from most network application services by decrypting the traffic. It is also difficult to classify various service flows through traditional network traffic classification methods into distinct application categories for management. This research assumes that Internet Service Providers (ISPs) have to manage tens of thousands of smart homes equipped with various kinds of IoT devices. We used software-defined networking (SDN) technology to simulate a multi-tenant smart home environment, simulate different types of smart home service traffic, and use convolutional neural networks (CNN) to classify network traffic. ISP operators can thus set the bandwidth ratio according to the classified service category to effectively improve QoS and user QoE. The experimental results show that the traffic classification accuracy of the CNN model for smart homes can reach 86.5%, which is higher than the general neural network model by 6.5%.