Currently, new multimedia applications are booming, especially streaming services represented by eXtended Reality (XR) and the next generation B5G/6G networks. Network traffic generated by XR and traditional mobile and PC clients is becoming more and more intertwined and complex, especially the homogeneity of Quality of Services (QoS) parameters makes the analysis of traffic critical. In this paper, we review the research done so far in traffic classification for XR services and propose the use of graph neural networks and Quality of Experience (QoE) scores to detect and classify highly similar streaming traffic as well as to predict them in further works. The goal will be not only to secure network resources and devices but also to achieve dynamic traffic classification and resource management for heterogeneous networks with a mixture of multiple network standards to meet the requirements of Self Organizing Networks (SON).