Traffic Flow Prediction using Spatiotemporal Fusion with Graph Convolution Network

Traffic flow prediction is one of the significant component of Intelligent Transportation System (ITS). However, higher-precision of traffic flow prediction particularly in long-term forecast is challenging because of complex spatial-temporal dependencies in traffic data. Therefore, this research proposes a Spatiotemporal Fusion with Graph Convolution Network (SF-GCN) for traffic flow prediction. The SF-GCN effectively capturing both spatial and temporal dependencies in traffic data thereby leveraging graph structures to represent spatial relationship accurately. Moreover, it integrates temporal dynamics that is significant in traffic prediction through spatiotemporal fusion which enables to understand changes in traffic flow patterns. The SF-GCN delivers accurate and robust traffic flow prediction through leveraging spatial-temporal complexity in traffic networks. The metrics such as Root mean square error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) are taken to estimate the SF-GCN performance in traffic flow prediction. The SF-GCN attains MAPE of $10.62 \%$ and $8.21 \%$ for PEMS04 and PEMS08 datasets which is better than Weather interaction-aware Spatio-Temporal Attention Network (WST-ANet).