A Smart Traffic Flow Optimization Using Graph Convolutional Network with Graph Long Short-Term Memory

In recent years, Smart Cities (SC), Traffic Flow (TF) refers to the optimized movement of vehicles, pedestrians, and other road users by progressive technologies. Predicting traffic flow is dynamic to ease congestion, enhance traffic management, and to improve emergency response. Previously, many methods were proposed to enhance the prediction of traffic flow, although the existing methods have overfitting, computational complexity, and generalizability problems. To overcome these problems, a Graph Convolutional Network based Graph Long Short-Term Memory (GCN-GLSTM) model is considered to predict traffic flow. Initially, Urban Traffic Density in Cities (UTDC) dataset is employed to optimize the flow of traffic. The preprocessing is performed with Data cleaning, to remove outliers and missing values, Data filtering to remove data points with low GPS accuracy, Z-Score normalization to normalize data, and Data transformation to Converted data into suitable formats. Feature extraction is performed by employing Graph Attention Network (GAN) to handle complexity. For classification, GCN-GLSTM is used to handle high-dimensionality, and to enhance generalizability. The proposed GCN-GLSTM method gained high performance metrics such as accuracy, MSE, MAE, and RMSE of 98.34%, 1.72, 1.14, and 1.35 compared to existing methods like Graph Neural Network (GNN).