Estimating travel time in transport network with a combined multi-attributed graph convolutional neural network and multilayer perceptron model

In this article, an advanced model for forecasting travel time in road networks is presented, employing a Graph Convolutional Neural Network (GCN) integrated with a Multilayer Perceptron (MLP), focusing on the travel time ratio (TTR). Utilizing data from Poland's Mazovia region, the model analyzes spatial and temporal dependencies in road infrastructure, incorporating road conditions, weather, and time for accurate traffic prediction. The research involved detailed analysis of neuron configurations and learning rates in the GCN-MLP model, focusing on their impact on mean squared error (MSE) and other key performance metrics. The optimized model configurations achieved a Mean Absolute Percentage Error (MAPE) of approximately 8%, outperforming many models in existing literature. The study found that hours, days, weather conditions, and technical road features significantly influence the model's outcomes. Efforts to optimize the model included excluding attributes with minimal impact on MSE, leading to notable improvements in efficiency. A comparison between the GCN-MLP model and traditional MLP models demonstrated the former's superior effectiveness, particularly in achieving lower MSE values and better predictive accuracy. These results highlight the potential and challenges of applying advanced machine learning techniques in traffic management, indicating a significant step forward in this field.