AMGST: Traffic Forecasting Performance Comparison Under Different K Values

AMGST, an Adaptive Multi-Graph Convolution and Spatiotemporal Multi-Head Self-Attention Network, demonstrates performance on four public traffic datasets. The 5.5 KB Excel file, authored by Pei Shi and last updated in June 2026, compares results under different K values. It includes speed and flow measurements from traffic datasets.

Use Cases

• Benchmarking graph-based neural network architectures for traffic forecasting based on the described AMGST model.
• Analyzing the impact of different K parameter values on model performance in spatiotemporal tasks.
• Comparing adaptive graph construction methods, such as maximum mutual information coefficient matrices, for capturing spatial correlations.
• Evaluating the integration of spatiotemporal attention mechanisms for modeling long-term dynamic dependencies in traffic patterns.

Strengths

• Results are validated on four public traffic datasets, as stated in the description.
• The dataset is small (5.5 KB), facilitating quick inspection and download.
• It is shared under a permissive CC-BY-4.0 license, allowing for broad reuse.

Limitations

• Row count and column-level documentation are unknown, which limits suitability assessment.
• The dataset's small size suggests it contains summary or comparison results, not the raw traffic data itself.
• Description metadata is limited; actual data quality and structure require manual inspection after download.

Provenance

Source

figshare

Collection Method

Likely contains experimental results from the AMGST model evaluation.

Freshness

Last updated 2026-06-04 17:36:41; freshness should be verified.