An Efficient and Effective Architecture for Large-Scale Traffic Prediction via Geometry-Adaptive Square Partitioning

2026-06-18

AI Summary

This work addresses the computational scalability limitations of existing neural network approaches in large-scale traffic forecasting and the suboptimal spatial partitioning induced by handcrafted geometric heuristics, which often yield imbalanced partitions and fragmented boundaries that degrade model accuracy. To overcome these issues, the authors propose the SqLinear architecture, which integrates a theoretically grounded, geometry-adaptive Square Partition algorithm—guaranteeing near-square, balanced, and high-utilization partitions—with a hierarchical linear interaction (HLI) module that models global and local spatiotemporal dynamics without attention mechanisms, achieving linear computational complexity. Empirical evaluations demonstrate that the proposed method reduces mean absolute error (MAE) by 2.30% under standard settings and by 5.81% under extreme scalability settings across four large-scale datasets, while also decreasing training time by 13.27%–30.84%.

Abstract

Traffic prediction is a core task in intelligent transportation systems and urban-scale decision making. Despite the effectiveness of mainstream neural-network based methods, their deployment in real-world settings with thousands of traffic sensors is jeopardized severely by their poor computational scalability. To address this, the community has attempted to incorporate spatial database partitioning techniques (e.g., Grid, Quadtree, and K-D Tree) to improve model scalability. However, these approaches rely on handcrafted geometric heuristics and often produce irregular or imbalanced data partitions, leading to boundary fragmentation, excessive padding overheads, and degraded model accuracy. In this paper, we propose SqLinear, an efficient and effective architecture for large-scale traffic prediction. First, we design Square Partition, a geometry-adaptive algorithm that partitions massive traffic sensors into balanced, non-overlapping, and near-square spatial regions. Unlike existing heuristic-based designs, Square Partition is theoretically grounded and provides provable guarantees on aspect ratio, balance, and partition utilization, establishing a high-quality foundation for downstream spatiotemporal modeling. Next, we propose a Hierarchical Linear Interaction (HLI) module that abandons the costly attention mechanisms commonly used in Transformer-based spatio-temporal models. HLI efficiently captures both local intra-region dynamics and global inter-region dependencies through a lightweight linear interaction scheme, enabling effective spatiotemporal modeling with linear computational complexity. Extensive experiments on four large-scale traffic datasets and 10 baselines show that SqLinear reduces MAE by 2.30% on average under standard setting and by 5.81% under extreme scalability settings, while reducing training runtime by 13.27%--30.84% in spatial- and horizon-scaling scenarios.

Problem

• traffic prediction
• computational scalability
• spatial partitioning
• model accuracy
• large-scale

Innovation

• Square Partition
• Hierarchical Linear Interaction
• Scalable Traffic Prediction
• Geometry-Adaptive Partitioning
• Linear Complexity