FHT-R: Federated Online Regression with Coordinated Hoeffding Trees for Air Quality Prediction

This paper presents a case study on federated online regression for PM2.5 prediction in distributed air quality monitoring. PM2.5 is a widely used air quality indicator that refers to fine particulate matter smaller than 2.5 µm. Originally, the Fed-VFDT framework was designed for classification. In this work, we adapt Fed-VFDT to regression using Hoeffding Tree regressors. Each monitoring station is modeled as a federated client that learns from its own local data stream, while a coordinating server selects global splits using quorum, majority vote, or best merit. Experiments on the Beijing Multi-Site Air Quality dataset, partitioned into 12 federated clients, showed that majority vote achieved the best predictive performance. These results indicate that coordinated federated tree growth is a feasible approach for distributed online air quality regression.