ISTGCN: Integrated spatio-temporal modeling for traffic prediction using traffic graph convolution network

To effectively estimate traffic patterns, spatial-temporal information must consider the complex spatial connections on road networks and time-dependent traffic information. Although deep learning models can comprehend the complex Spatio-temporal correlations in traffic data, much research has been done recently on creating these models for traffic prediction. Aside from graph neural networks (GNNs), most contemporary approaches have created spatial-only and temporal-only units to learn the spatial and temporal characteristics separately. The physical network topology is used to explain the traffic graph convolution. Therefore, this paper proposed an Integrated Spatio-Temporal Graph Convolution Network for traffic prediction (ISTGCN), which learns the accumulation of temporal and spatial properties by transmitting data from various timestamps junctions. The aim is to propose a solution without using separate Spatio-temporal modules. Experiments show that the proposed ISTGCN model effectively captures spatial-temporal information and improves traffic prediction accuracy with the state-of-the-art model. The proposed model compared on four windows, prediction after 15, 30, 45 and 60 minutes. ISTGCN improved the MAE, RMSE, and MAPE by 20%, 6.11% and 18%, respectively of USTGCN, while predicting after 60 minutes. The results also show that the training time is substantially reduced due to the smaller number of training parameters.