Spectral method improves long-term traffic forecasting accuracy
SpecLST: Spectral-Enhanced Long-Short Term Spatio-Temporal Multimodal Modeling for Traffic Forecasting
Machine Learning
Summary
Accurately predicting traffic over long periods is hard because it requires analyzing lots of past data and combining different types of information like incidents and weather. The authors created SpecLST, a new approach that uses frequency techniques to efficiently capture long-term traffic patterns without heavy computing costs. SpecLST also better detects sudden changes in traffic by focusing on quick shifts captured by high-frequency data. Tests show it reduces prediction errors significantly compared to previous methods.
What this means in practice
- •For urban traffic planners: Improve traffic management systems by providing more accurate and longer-term traffic flow predictions using spectral-enhanced models.
- •For logistics and delivery coordinators: Use forecasts that capture sudden traffic incidents and weather impacts more precisely to optimize route planning and delivery schedules.
Authors
Kenan Kang, Matthew M. Y. Kuo, Weihua Li
Abstract
Modeling multi-scale dynamics in urban road networks is essential for intelligent transportation systems but is often hindered by the computational cost of long-range historical context and the challenge of fusing heterogeneous modalities. Existing spatio-temporal frameworks face a trade-off between sequence length and memory overhead, while frequently neglecting the transient effects of exogenous factors such as road incidents and weather. To address these limitations, we propose SpecLST, a frequency-enhanced long-short-term spatio-temporal multimodal framework. SpecLST leverages a fast Fourier transform-based frequency-domain representation module to efficiently encode multi-week historical dependencies, bypassing the memory bottlenecks of standard attention mechanisms over raw long sequences. Heterogeneous context is integrated through attention-based fusion, followed by a sparse Mixture-of-Experts backbone with graph augmentation for spatio-temporal representation learning. Additionally, a frequency-enhanced residual head isolates high-frequency transients, improving sensitivity to abrupt traffic anomalies. At the 60-minute horizon, SpecLST achieves mean relative error reductions of 14.10% in MAE, 10.35% in RMSE, and 15.78% in MAPE across eight neural forecasting baselines and four datasets, with the dataset- and metric-specific average reductions ranging from 5.63% to 22.21%.