Adaptive Multi-Scale Graph Neural Network for Dynamic Spatial-Temporal Traffic Flow Forecasting

Abstract

This dataset supports the paper “MGraph: Adaptive Multi-Graph Neural Network for Dynamic Spatial-Temporal Traffic Flow Forecasting,” published in IEEE Transactions on Intelligent Transportation Systems. It contains experimental results for the MGraph model, which predicts traffic flow using adaptive adjacency matrices and a weekly regularity scheme. The dataset includes txt files with predictions, ground truth, and evaluation metrics (MAE, RMSE, MAPE) for four traffic datasets (PeMS03, PeMS04, PeMS07, PeMS08) across four prediction horizons (15, 30, 45, 60 minutes). Organized by model, dataset, and random seed (1–10), the files enable reproducibility of MGraph’s performance and benchmarking against baselines like GWNet and STID. Researchers can use this dataset to validate results, compare new models, or analyze traffic patterns. Code is available at https://github.com/JABUBROWN/MGraph.

Instructions:

## OverviewThis dataset supports the paper "MGraph: Adaptive Multi-Graph Neural Network for Dynamic Spatial-Temporal Traffic Flow Forecasting," submitted to IEEE Transactions on Intelligent Transportation Systems (2025). It contains experimental results for the MGraph model and 10 baseline models, evaluated on four traffic datasets (PeMS03, PeMS04, PeMS07, PeMS08). The dataset includes raw result files (predictions and metrics) and analysis summaries, enabling reproducibility of the paper’s findings (Tables 1 and 2) and benchmarking for traffic forecasting research.

## Dataset Description

The dataset is stored in the Final_Experiments-NEW-NEW/ directory, which contains two primary components:

1. Ablation/ – Results and statistical summaries for the MGraph ablation study

         - results/ — 80 raw experiment files (2 ablation variants × 4 datasets × 10 seeds)

         - analysis/ — 2 aggregated summary files, each reporting mean and standard deviation across 10 seeds

2. Main/ – Full benchmarking results for all baseline and comparison models

         - Results/ — 440 raw experiment files (11 models × 4 datasets × 10 seeds)

         - Analysis/ — 11 aggregated summary files, one per model

In total, the dataset contains 520 raw .txt result files and 13 statistical summary files. Every model–dataset pair is evaluated using 10 independent random seeds to ensure reproducibility and statistically reliable reporting.

### Datasets- **PeMS03**: 358 sensors, 26,208 time steps, regional highway traffic.- **PeMS04**: 307 sensors, 16,992 time steps, regional highway traffic.- **PeMS07**: 883 sensors, 28,224 time steps, regional highway traffic.- **PeMS08**: 170 sensors, 17,856 time steps, regional highway traffic.

### Models- **Main Experiments**: MGraph, AGCRN, ASTGCN, DSTAGNN, FourierGNN, GMAN, GWNet, SCINet, STGCN, STID, STSGCN (11 models).- **Ablation Studies**: - Ablation1: MGraph without weekly regularity scheme. - Ablation2: MGraph without adaptive adjacency matrix.

### Metrics- **MAE**: Mean Absolute Error.- **RMSE**: Root Mean Square Error.- **MAPE**: Mean Absolute Percentage Error (%).- Evaluated at 12 prediction horizons: 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60 minutes.Final_Experiments/├── Ablation│   ├── analysis│   │   ├── Ablation1_analysis.txt│   │   └── Ablation2_analysis.txt│   └── results│       ├── Ablation1│       │   ├── ablation1_MGraph_PeMS03_seed10.txt│       │   ├── ablation1_MGraph_PeMS03_seed1.txt│       │   ├── ablation1_MGraph_PeMS03_seed2.txt│       │   ├── ...│       └── Ablation2│           ├── ablation2_MGraph_PeMS03_seed10.txt│           ├── ablation2_MGraph_PeMS03_seed1.txt│           ├── ablation2_MGraph_PeMS03_seed2.txt│           ├── ablation2_MGraph_PeMS03_seed3.txt│           ├── ...└── Main    ├── Analysis    │   ├── AGCRN_analysis.txt    │   ├── ASTGCN_analysis.txt    │   ├── DSTAGNN_analysis.txt    │   ├── FourierGNN_analysis.txt    │   ├── GMAN_analysis.txt    │   ├── GWNet_analysis.txt    │   ├── MGraph_analysis.txt    │   ├── SCINet_analysis.txt    │   ├── STGCN_analysis.txt    │   ├── STID_analysis.txt    │   └── STSGCN_analysis.txt    └── Results        ├── AGCRN_PeMS03_seed10.txt        ├── AGCRN_PeMS03_seed1.txt        ├── AGCRN_PeMS03_seed2.txt        ├── AGCRN_PeMS03_seed3.txt        ├── AGCRN_PeMS03_seed4.txt        ├── AGCRN_PeMS03_seed5.txt        ├── AGCRN_PeMS03_seed6.txt        ├── AGCRN_PeMS03_seed7.txt        ├── AGCRN_PeMS03_seed8.txt        ├── ...