Taxi Traffic Benchmark

Pipeline

Dataset

• Shanghai Taxi
• Note: This link has expired. The processed data is located in the "/data" directory.

Baseline

• ARIMA
• SVR
• LSTM
• STGCN[Paper] [Code] [Official Code]
• ASTGCN[Paper] [Code] [Official Code]
• Graph WaveNet[Paper] [Code] [Official Code]

Proposed

[Paper]MA2GCN: Multi Adjacency relationship Attention Graph Convolutional Networks for Traffic Prediction using Trajectory data

Architecture	Multi Adjacency relationship Attention mechanism

Related Links

• GPS2Graph Link

Important Notes

• The Shanghai Taxi Dataset rarely appears in traffic prediction tasks, so the six baselines above can be used as a reference.
• The 'io_matrix' in code corresponds to the 'vehicle entry and exit matrix ' in ma2gcn paper. The 'io_adj' in code corresponds to the $A_{mo}$ in ma2gcn paper.
• The six baselines can't use the mobility matrix as it's specificly used for proposed ma2gcn model.
• Due to the mobility matrix is time sensitive, the dataset needs to be divided in chronological order and cannot be shuffled. For the six baselines, there is no such limit.
• The proposed model is on arxiv.org. Now it can be seen as a simple technical report exploring the use of taxi trajectories for traffic prediction tasks. There's still significant potential for enhancement.