Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Strada-LLM: Graph LLM for traffic prediction

Seyed Mohamad Moghadas, Bruno Cornelis, Alexandre Alahi, Adrian Munteanu

TL;DR

This work introduces Strada-LLM, a novel multivariate probabilistic forecasting LLM that explicitly models both temporal and spatial traffic patterns, and proposes a lightweight distribution-derived strategy for domain adaptation to enhance adaptability and interpretability in real-world traffic networks.

Abstract

Traffic forecasting is pivotal for intelligent transportation systems, where accurate and interpretable predictions can significantly enhance operational efficiency and safety. A key challenge stems from the heterogeneity of traffic conditions across diverse locations, leading to highly varied traffic data distributions. Large language models (LLMs) show exceptional promise for few-shot learning in such dynamic and data-sparse scenarios. However, existing LLM-based solutions often rely on prompt-tuning, which can struggle to fully capture complex graph relationships and spatiotemporal dependencies-thereby limiting adaptability and interpretability in real-world traffic networks. We address these gaps by introducing Strada-LLM, a novel multivariate probabilistic forecasting LLM that explicitly models both temporal and spatial traffic patterns. By incorporating proximal traffic information as covariates, Strada-LLM more effectively captures local variations and outperforms prompt-based existing LLMs. To further enhance adaptability, we propose a lightweight distribution-derived strategy for domain adaptation, enabling parameter-efficient model updates when encountering new data distributions or altered network topologies-even under few-shot constraints. Empirical evaluations on spatio-temporal transportation datasets demonstrate that Strada-LLM consistently surpasses state-of-the-art LLM-driven and traditional GNN-based predictors. Specifically, it improves long-term forecasting by 17% in RMSE error and 16% more efficiency. Moreover, it maintains robust performance across different LLM backbones with minimal degradation, making it a versatile and powerful solution for real-world traffic prediction tasks.

Strada-LLM: Graph LLM for traffic prediction

TL;DR

This work introduces Strada-LLM, a novel multivariate probabilistic forecasting LLM that explicitly models both temporal and spatial traffic patterns, and proposes a lightweight distribution-derived strategy for domain adaptation to enhance adaptability and interpretability in real-world traffic networks.

Abstract

Traffic forecasting is pivotal for intelligent transportation systems, where accurate and interpretable predictions can significantly enhance operational efficiency and safety. A key challenge stems from the heterogeneity of traffic conditions across diverse locations, leading to highly varied traffic data distributions. Large language models (LLMs) show exceptional promise for few-shot learning in such dynamic and data-sparse scenarios. However, existing LLM-based solutions often rely on prompt-tuning, which can struggle to fully capture complex graph relationships and spatiotemporal dependencies-thereby limiting adaptability and interpretability in real-world traffic networks. We address these gaps by introducing Strada-LLM, a novel multivariate probabilistic forecasting LLM that explicitly models both temporal and spatial traffic patterns. By incorporating proximal traffic information as covariates, Strada-LLM more effectively captures local variations and outperforms prompt-based existing LLMs. To further enhance adaptability, we propose a lightweight distribution-derived strategy for domain adaptation, enabling parameter-efficient model updates when encountering new data distributions or altered network topologies-even under few-shot constraints. Empirical evaluations on spatio-temporal transportation datasets demonstrate that Strada-LLM consistently surpasses state-of-the-art LLM-driven and traditional GNN-based predictors. Specifically, it improves long-term forecasting by 17% in RMSE error and 16% more efficiency. Moreover, it maintains robust performance across different LLM backbones with minimal degradation, making it a versatile and powerful solution for real-world traffic prediction tasks.

Paper Structure

This paper contains 27 sections, 7 equations, 6 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Methodology
  4. Hierarchical Feature Extractor
  5. $k$-hop Sub-Graph Extractor:
  6. Global Graph Embedding
  7. Lag Extractor
  8. LLM-based-Backbone
  9. Domain Adaptation
  10. Low-Rank Matrix Adaptation
  11. Loss Function
  12. EXPERIMENTS
  13. Datasets Details
  14. Baselines
  15. LLM Forecasting Results
  16. ...and 12 more sections

Figures (6)

  • Figure 1: An example of the spatio-temporal dependency between the traffic flows for different traffic states ①,②,③. Nodes (a), (b), and (c) are placed in a crowded area of Brussels. The corresponding traffic patterns between 8:00 am and 8:00 pm are illustrated for nodes (a), (b), and (c). The short-term traffic flow in (c) is altered by the significant impact of neighboring roads, resulting in varying flows.
  • Figure 2: The proposed Strada-LLM architecture. The core novelty lies in the K-hop sub-graph extractor module, which takes lag features in both the spatial and temporal dimensions.
  • Figure 3: Conceptual LLM adaptation approaches comparison
  • Figure 4: Comparison of unified benchmark for different LLMs.
  • Figure 5: Zero-shot performance comparison on Crowd dataset.
  • ...and 1 more figures