Traffic Flow Prediction with Vehicle Trajectories

Authors: Mingqian Li, Panrong Tong, Mo Li, Zhongming Jin, Jianqiang Huang, Xian-Sheng Hua294-302

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed approach is evaluated with a real-world trajectory dataset. Experiment results show that the proposed Tr GNN model achieves over 5% error reduction when compared with the state-of-the-art approaches across all metrics for normal traffic, and up to 14% for atypical traffic during peak hours or abnormal events.
Researcher Affiliation Collaboration 1 Alibaba-NTU Singapore Joint Research Institute, Nanyang Technological University, Singapore 2 School of Computer Science and Engineering, Nanyang Technological University, Singapore 3 Alibaba Group
Pseudocode No The paper describes the model architecture and components in text and diagrams but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Code and dummy data are available at https://github.com/mingqian000/Tr GNN.
Open Datasets No The paper describes the SG-TAXI dataset and its source (Singapore Land Transport Authority), but it does not provide a public link, DOI, or specific citation for direct access to the dataset, which is necessary for reproducibility.
Dataset Splits Yes The train-validate-test split is 5-1-2 week.
Hardware Specification Yes The model is implemented in Py Torch (Paszke et al. 2019) on a single Tesla P100 GPU
Software Dependencies No The paper mentions that the model is implemented in PyTorch, but it does not specify a version number for PyTorch or any other software dependency for its own implementation. It mentions specific versions for some baselines (e.g., TensorFlow 2), but not for the proposed Tr GNN model.
Experiment Setup Yes For hyperparameters, the demand hop d is set to 75, i.e., the maximum number of road segments that a vehicle with a normal speed could traverse within a 15-minute interval, and the status hop s is set to 3. The learning rate is initially set to 0.004 and is halved every 30 epochs. The maximum epochs to train is set to 100. Early stopping is applied on validation MAE.