Scholarlink

This research investigates the efficacy of various Graph Neural Network (GNN) model architectures in traffic prediction tasks. We compare standard GNN models with two variants: one excluding edge features and another utilizing a k-nearest neighbors graph structure. Our comprehensive comparative analysis examines prediction accuracy and computational efficiency across these models. Experimental results demonstrate that GNN models significantly outperform the Baseline model in both prediction accuracy and training efficiency. Among the models tested, the GNN architecture incorporating edge, node, and global models exhibits superior performance, highlighting the critical role of graph structures in enhancing traffic prediction capabilities. The study underscores the potential of GNNs to improve predictive modeling in traffic systems by leveraging their ability to capture complex relationships within the data.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Traffic Forecasting with Spatio-Temporal Graph Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Shehal Shah,
  • Prince Doshi,
  • Shlok Mangle,
  • Prachi Tawde,
  • Vinaya Sawant

摘要

This research investigates the efficacy of various Graph Neural Network (GNN) model architectures in traffic prediction tasks. We compare standard GNN models with two variants: one excluding edge features and another utilizing a k-nearest neighbors graph structure. Our comprehensive comparative analysis examines prediction accuracy and computational efficiency across these models. Experimental results demonstrate that GNN models significantly outperform the Baseline model in both prediction accuracy and training efficiency. Among the models tested, the GNN architecture incorporating edge, node, and global models exhibits superior performance, highlighting the critical role of graph structures in enhancing traffic prediction capabilities. The study underscores the potential of GNNs to improve predictive modeling in traffic systems by leveraging their ability to capture complex relationships within the data.