Traffic Forecasting using Graph Convolution Network

Nowadays, accurately forecasting traffic in urban road networks is essential in intelligent traffic systems. Therefore, this work aims to overcome the problem of spatial and temporal dependence, which occurs in most traffic forecasting models. Graph Convolutional Network (GCN) helps find the spatial dependence, whereas Gated Recurrent Unit (GRU) is used to analyze the dynamic change in traffic, i.e., capture temporal dependence. In this paper, a combined model using GCN and GRU is proposed to address the traffic forecasting problem in the urban environment. The proposed model is named as Spatio-Temporal Graph Convolutional Network (ST-GCN) predicts the traffic. It can be concluded from the results that the ST-GCN model gives better accuracy than the other available models.