Granular network traffic classification is gaining high priority, which is crucial for the Internet Service Providers (ISP), Over The Top (OTT) providers' operation and maintenance management and users' Quality of Experience (QoE) improvement. Streaming video category traffic takes up a significant proportion of Internet traffic. Its ground truth sources include the application types, streaming network communication protocols, resolution, refresh rates, video encoding protocols, physical network resource bandwidth, and associated with the video source. However, the credibility of the traffic classification work based on the ground truths above-mentioned is not high since the quality of the video source cannot be guaranteed. The user's perception is poor even when watching higher resolution and refresh rate in a particular scenario. Secondly, different video platforms use different technical standards, which will inevitably cause video quality compression loss in transmission and viewing. The user viewing experience varies greatly, even under the same standard. In this paper, we propose to implement the traffic classification task by calculating the video bitrates of Video on Demand (VoD) and Video Live Streaming (VLS) as accurate classification labels and using machine learning techniques, in which we will examine the real-time bitrates during real-world video transmission compared with the bitrates set by theoretical recommendations to find the differences between the two scenarios.