Efficient Traffic Prediction at Scale: A Systematic Study of STGCN Architectural Depth

2026-06-08

AI Summary

This study addresses the high computational cost of existing spatio-temporal graph neural networks, which hinders their deployment in resource-constrained intelligent transportation systems. The authors systematically evaluate the performance–efficiency trade-offs of Spatio-Temporal Graph Convolutional Networks (STGCNs) with depths ranging from one to three blocks, revealing for the first time that the standard two-block STGCN is significantly over-parameterized. Experimental results demonstrate that a single-block architecture achieves optimal or near-optimal short-term prediction accuracy across three widely used traffic datasets, while reducing inference latency by 61% and increasing throughput by 37%. In contrast, extending the model to three blocks yields negligible performance gains at the cost of nearly doubling computational overhead. These findings provide critical insights for designing lightweight yet effective traffic forecasting models.

Abstract

Spatio-temporal graph neural networks (STGNNs) have become the dominant approach for traffic prediction, yet their computational requirements pose challenges for practical deployment in intelligent transportation systems (ITS). While recent work has proposed efficient alternatives to STGNNs, a fundamental question remains unexplored: are these architectures themselves over-parameterised? We examine this question using the Spatio-Temporal Graph Convolutional Network (STGCN), one of the most widely adopted models in this domain. Through systematic experiments across four diverse traffic datasets, we compare 1-block, 2-block (standard), and 3-block STGCN variants. Our findings reveal that the single-block architecture achieves optimal performance for short-term prediction (10 mins) on three of four datasets, while incurring only marginal degradation ($\leq$1.8% relative error) at longer horizons. Crucially, the 2-block variant incurs 61% higher CPU inference latency and 37% lower throughput relative to 1-block -- substantial overhead for resource-constrained ITS deployment. The 3-block architecture offers no favourable tradeoff, more than doubling computational cost for $<$0.5% relative improvement. These results suggest that the default 2-block STGCN may be over-parameterised for many applications, with implications for both practitioners deploying traffic prediction systems and researchers benchmarking efficiency-focused methods.

Problem

• traffic prediction
• spatio-temporal graph neural networks
• model over-parameterization
• computational efficiency
• intelligent transportation systems

Innovation

• STGCN
• over-parameterisation
• traffic prediction
• computational efficiency
• spatio-temporal graph neural networks