Artificial Intelligence-Enabled Data Forecasting Methodology for Smart Transportation System

Since traffic patterns alter over time and have complicated spatial connections owing to road systems, traffic forecasting provides unique problems for spatiotemporal forecasting. To address this challenge, we introduce the Graph-based Gated Recurrent Neural Network (G2RNN), a novel deep learning framework that uses graphs to learn the interactions between roads and make network-wide traffic predictions. Our concept of the flow graph inversion is based on the physical network architecture. We also elaborate on the connection between our proposed traffic graph convolution and the spectral graph convolution. G2RNN features are added to a loss function to increase the model's interpretability. Experimental findings show that the suggested approach outperforms baseline techniques on two traffic in the real world state datasets. Graph convolution weights visualisation demonstrates the proposed system's ability to pinpoint critical nodes in actual traffic networks.