Some Commonly Used Features are Not Reliable: Enhancing Command Identification in Smart Home Wi-Fi Traffic Data
Pingchuan Ma, Zexiong Xiang, Yiwen Li, Jian Chen, Shanshan Wang, Zhenxiang Chen · 2025
Nowadays, people are accustomed to using various smart home devices, which generate a lot of network traffic data every day. Accurately identifying user commands in traffic helps detect privacy leaks and potential security risks, leading to more effective management and maintenance of IoT devices. As a result, this issue has become a key focus for researchers in recent years. However, based on the observation of real data, we found that two categories of important features commonly used in current research-time and packet length features-have problems with insufficient representation and differentiation when used for user command identification. Then based on the findings, this paper proposes a new representation method for the command traffic. This method can amplify the distinction between command traffic and also encode time and packet length information. In addition, we established a real smart home environment and conducted experiments. The results show that the method in this paper can achieve command identification with an accuracy of 91%, surpassing existing methods.