PeMSD4 and PeMSD8: California Traffic Flow Data for Spatiotemporal Modeling

Two California traffic datasets, PeMSD4 and PeMSD8, used to benchmark a Transformer-based Hypergraph Convolutional Network for traffic flow prediction. The 5.5 KB XLS files were published by SiWei Wei on figshare under a CC-BY-4.0 license and last updated in April 2026. Experiments described in the metadata involved 5 independent random seed runs to achieve statistically significant performance improvements on core metrics like MAE, RMSE, and MAPE.

Use Cases

• Benchmarking traffic flow prediction models based on MAE, RMSE, and MAPE metrics.
• Training models to capture high-order nonlinear spatial correlations between traffic nodes.
• Developing spatiotemporal feature fusion mechanisms for intelligent transportation systems.

Strengths

• Dataset is small at 5.5 KB, facilitating quick download and initial inspection.
• License is permissive (CC-BY-4.0), allowing for broad reuse and modification.
• Metadata includes a detailed description of the experimental methodology used with the data.

Limitations

• Row count is unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• The description focuses on a model experiment; the raw data's structure and scope are not detailed.

Provenance

Source

figshare, author SiWei Wei.

Collection Method

Likely sourced from the California Performance Measurement System (PeMS).

Time Range

null

Freshness

Last updated 2026-04-28 17:48:24; freshness should be verified.

Geography

California, based on the dataset names and description.