DSTGN-ExpertNet: A Deep Spatio-Temporal Graph Neural Network for High-Precision Traffic Forecasting

Accurate traffic prediction is essential for optimizing urban mobility and mitigating congestion.Traditional deep learning models, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), struggle to capture complex spatiotemporal dependencies and dynamic traffic variations across urban networks.To address these challenges, this study introduces DSTGN-ExpertNet, a novel Deep Spatio-Temporal Graph Neural Network (DSTGNN) framework that integrates Graph Neural Networks (GNNs) for spatial modeling and advanced deep learning techniques for temporal dynamics.The framework employs a Mixture of Experts (MoE) approach, where specialized expert models are dynamically assigned to distinct traffic patterns through a gating network, optimizing both prediction accuracy and interpretability.The proposed model is evaluated on large-scale realworld traffic datasets from Beijing and New York, demonstrating superior performance over conventional methods, including Spatio-Temporal Graph Convolutional Networks (ST-GCN) and attention-based models.With a mean absolute error (MAE) of 1.97 on the BikeNYC dataset and 9.70 on the TaxiBJ dataset, DSTGN-ExpertNet achieves state-of-the-art accuracy.These findings highlight the potential of GNN-based frameworks in revolutionizing traffic forecasting and intelligent transportation systems (ITS).