Los Angeles Metropolitan Traffic Forecasting using Graph Neural Network

Traffic forecasting is critical for urban transportation management, particularly in large metropolitan areas where congestion is a persistent issue. In this paper, we propose a traffic forecasting model based on Graph Convolutional Networks (GCNs) to capture both spatial and temporal dependencies in traffic data. We utilize a dataset consisting of traffic flow information from 207 loop detectors deployed across the Los Angeles metropolitan highways, covering the period from March 2012 to June 2012. The proposed model, A3<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$T$</tex>-GCN, is evaluated against three benchmark models: a standard Graph Neural Network (GNN), Long Short-Term Memory network (LSTM), and a simple Neural Network (NN). Our experimental results demonstrate that GCN-based models significantly outperform traditional methods. Specifically, the GNN achieves the lowest Mean Squared Error (MSE) Loss of 0.762, followed closely by A3T-GCN with 0.771. In contrast, the LSTM and NN models exhibit higher MSE losses of 0.835 and 0.848, respectively. These results highlight the effectiveness of GCNs in addressing the spatial-temporal complexities inherent in traffic forecasting tasks for urban planning.