Urban traffic forecasting using attention based model with GCN and GRU.

Now-a-days since number of vehicles are growing day by day on the roads in urban and metropolitan area, hence traffic jam is becoming very common problem which everyone is facing time to time. So it becomes very crucial to forecast traffic to avoid the regular traffic congestion situation in daily life. In today's intelligent traffic system, reliable and precise traffic forecasting in metropolitan road networks is critical. Many models based on classical methods have been presented to address this; however, they lag in certain circumstances, failing to incorporate spatial and temporal dependency in the environment. As a result, this research aims to solve the geographical and temporal dependency problem that plagues most traffic forecasting models. Graph Convolution Network (GCN) and Gated Recurrent Unit (GRU) are employed in conjunction with an attention-based model for more accurate traffic forecasts in cities. A combined model called Spatio-Temporal Attention-based Model with GCN and GRU (ST AGG) is employed to anticipate traffic. ST-AGG model mainly focuses on extracting the relevant information from all the input given to the attention-based model. The spatial dependency is determined by using GCN and the temporal dependency is determined by using GRU, used in the ST-AGG model. It assists in short-term traffic forecasting as well as long-term traffic forecasting. The results demonstrate that the suggested ST AGG has properly predicted the volume of traffic on city roadways in real time with greater accuracy. The model depicts how traffic is influenced by geography and time. The Shen Zhen (SZ) car dataset from China and the Los Angeles (LAS) dataset from California are used to test the suggested model. And the result shows that Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) have reduced by up to 2.7% less, and accuracy has increased by 2.7% more compared to the previous model ST GCN for the SZ vehicle dataset. Similarly, although the accuracy has not changed significantly for the Loss Angeles data set, RMSE and MAE have reduced by 3.71% less as compared to the previous ST GCN model. The proposed model is more effective than the prior models like GCN, GRU, and ST GCN. [ABSTRACT FROM AUTHOR]