Adaptive Graph Pruning with Sudden-Events Evaluation for Traffic Prediction using Online Semi-Decentralized ST-GNNs

TL;DR

The paper tackles the high communication costs of online semi-decentralized ST-GNN training for traffic prediction by introducing an adaptive cross-cloudlet pruning algorithm and a new event-centric metric, SEPA. SEPA quantifies a model’s ability to detect sudden traffic slowdowns and recoveries, revealing benefits of cross-cloudlet connectivity that standard metrics miss, especially at longer horizons. The authors demonstrate that pruning boundary features adaptively maintains predictive accuracy while substantially reducing inter-cloudlet data transfer across traditional FL, server-free FL, and Gossip Learning setups on PeMS-BAY and PeMSD7-M datasets. Overall, the work provides a practical, scalable approach to edge-enabled traffic forecasting with improved responsiveness to critical events and clear guidance for future cloudlet-aware distributed learning strategies.

Abstract

Spatio-Temporal Graph Neural Networks (ST-GNNs) are well-suited for processing high-frequency data streams from geographically distributed sensors in smart mobility systems. However, their deployment at the edge across distributed compute nodes (cloudlets) createssubstantial communication overhead due to repeated transmission of overlapping node features between neighbouring cloudlets. To address this, we propose an adaptive pruning algorithm that dynamically filters redundant neighbour features while preserving the most informative spatial context for prediction. The algorithm adjusts pruning rates based on recent model performance, allowing each cloudlet to focus on regions experiencing traffic changes without compromising accuracy. Additionally, we introduce the Sudden Event Prediction Accuracy (SEPA), a novel event-focused metric designed to measure responsiveness to traffic slowdowns and recoveries, which are often missed by standard error metrics. We evaluate our approach in an online semi-decentralized setting with traditional FL, server-free FL, and Gossip Learning on two large-scale traffic datasets, PeMS-BAY and PeMSD7-M, across short-, mid-, and long-term prediction horizons. Experiments show that, in contrast to standard metrics, SEPA exposes the true value of spatial connectivity in predicting dynamic and irregular traffic. Our adaptive pruning algorithm maintains prediction accuracy while significantly lowering communication cost in all online semi-decentralized settings, demonstrating that communication can be reduced without compromising responsiveness to critical traffic events.

Figures (6)

• Figure 1: Graph partitioning and communication. a) Geographically distributed sensor network and base stations. b) Graph partitioning of the sensors into cloudlets based on geographical proximity. c) Cloudlet-to-cloudlet communication network for exchanging node features and model updates. d) After communicating with neighbouring cloudlets, each cloudlet can construct the ST-GNN subgraph required for training on its local nodes related_work_12.
• Figure 2: High-level flow of the Adaptive Cross-Cloudlet Pruning Algorithm for online semi-decentralized ST-GNN training
• Figure 3: Sensor assignment to cloudlets based on communication range
• Figure 4: Difference in absolute error between full and no graph connectivity for long-term prediction on the PeMS-BAY dataset (sensor 134)
• Figure 5: Communication cost, WMAPE, and SEPA for PeMS-BAY dataset and for four representative cloudlets (3, 4, 5, 7) under Server-free FL with 140-data timewindow (Left: short-term prediction; right: long-term prediction)