Two-Phase Traffic Intelligence for 6G SAGIN: Bayesian QoS Estimation and SLA-Conscious Slice Selection

The sixth generation (6G) of mobile networks is envisioned to support ultra-diverse services with highly heterogeneous Quality-of-Service (QoS) requirements, ranging from ultra-low latency and high reliability to massive connectivity and energy efficiency. Efficient traffic management and adaptive slice selection are therefore critical, particularly within the multi-layered and dynamic Space-Air-Ground Integrated Network (SAGIN) architecture. This paper proposes a two-phase classification and decision framework under a Knowledge-Defined Networking (KDN) paradigm to enable QoS-aware network intelligence in 6G SAGIN. In the first phase, incoming traffic flows are classified according to their latent QoS demands using a Bayesian probabilistic inference model over multidimensional flow-level features. In the second phase, the most suitable network slice is selected by matching the inferred QoS profile with slice-specific Service Level Agreements (SLAs) through an optimisation-based compatibility function. The proposed model is fully integrated into the KDN Knowledge Plane, enabling uncertainty-aware, real-time decision-making. Simulation results demonstrate that the framework significantly improves slice utilisation efficiency, SLA satisfaction, and end-to-end performance metrics, including delay and packet loss.