FedQoS Predicts QoS Risk for Wireless Access Selection

Ensuring reliable wireless access in dynamic indoor-outdoor environments is challenging because instantaneous signal measurements don't predict future Quality of Service (QoS) issues caused by mobility, blockage, or traffic. This research introduces FedQoS, a federated learning framework designed to predict QoS risk for better access node selection. FedQoS allows each access node to learn from its local network data, including radio conditions, traffic, and service context, while contributing to a globally trained QoS-risk predictor. The key advantage is that FedQoS achieves this without centralizing sensitive user-level network data, maintaining privacy. The learned model estimates the probability of QoS failure for candidate links, enabling controllers to select the most reliable access nodes. Simulations using synthetic indoor-outdoor datasets demonstrated that FedQoS substantially reduces QoS failure rates compared to signal-based or historical heuristic methods, achieving near-centralized performance even with non-IID data.