FUSE-Traffic: Fusion of Unstructured and Structured Data for Event-aware Traffic Forecasting

TL;DR

FUSE-Traffic addresses the challenge of predicting traffic under non-periodic, event-driven disruptions by integrating dynamically retrieved event information from large language models with a spatio-temporal graph neural encoder. The framework fuses event semantics and traffic dynamics through a cross-attention mechanism and a frozen text encoder, enabling on-the-fly adaptation to incidents, weather, and public activities. Across METR-LA, PEMS-BAY, and PEMS03, it outperforms eight baselines on MAE, RMSE, and MAPE and demonstrates robustness under varying event intensities, with ablations highlighting the importance of weather/crime signals and the fusion design. This work advances practical, robust traffic forecasting by leveraging LLM-based event querying to enrich structured forecasts, with potential for real-world ITS deployments.

Abstract

Accurate traffic forecasting is a core technology for building Intelligent Transportation Systems (ITS), enabling better urban resource allocation and improved travel experiences. With growing urbanization, traffic congestion has intensified, highlighting the need for reliable and responsive forecasting models. In recent years, deep learning, particularly Graph Neural Networks (GNNs), has emerged as the mainstream paradigm in traffic forecasting. GNNs can effectively capture complex spatial dependencies in road network topology and dynamic temporal evolution patterns in traffic flow data. Foundational models such as STGCN and GraphWaveNet, along with more recent developments including STWave and D2STGNN, have achieved impressive performance on standard traffic datasets. These approaches incorporate sophisticated graph convolutional structures and temporal modeling mechanisms, demonstrating particular effectiveness in capturing and forecasting traffic patterns characterized by periodic regularities. To address this challenge, researchers have explored various ways to incorporate event information. Early attempts primarily relied on manually engineered event features. For instance, some approaches introduced manually defined incident effect scores or constructed specific subgraphs for different event-induced traffic conditions. While these methods somewhat enhance responsiveness to specific events, their core drawback lies in a heavy reliance on domain experts' prior knowledge, making generalization to diverse and complex unknown events difficult, and low-dimensional manual features often lead to the loss of rich semantic details.

Figures (11)

• Figure 1: Event Impact on Traffic Status in PEMS-BAY
• Figure 2: Overview of Four Representative Fusion Architecture. Top to Bottom: a) STORM Hong2024STORM uses graph encoder to process event data and performs fusion & Alignment between two modalities. b) T3 EventTraffic2024 performs fusion by concatenating graph and text representations at the output stage. c) UrbanGPT Li2024UrbanGPT incorporates spatiotemporal data and regional location information into the prompt, which is then fed into an LLM for prediction. d) FUSE-Traffic utilizes LLM for query events and jointly processes both graph and text representations via fusion and alignment
• Figure 3: FUSE-Traffic Framework
• Figure 4: Distribution of Event Impact Type in 3 Datasets
• Figure 5: Performance Comparison in METR-LA

Theorems & Definitions (4)

• Definition 1: Road Network
• Definition 2: Traffic Status
• Definition 3: Event
• Definition 4: Event Aware Traffic Forecasting