Multi-Model Traffic Forecasting in Smart Cities using Graph Neural Networks and Transformer-based Multi-Source Visual Fusion for Intelligent Transportation Management

In the intelligent transportation management of smart cities, traffic forecasting is crucial. The optimization of traffic flow, reduction of congestion, and improvement of the overall transportation system efficiency all depend on accurate traffic pattern projections. In order to overcome the difficulties caused by the complexity and diversity of urban traffic dynamics, this research suggests a unique method for multi-modal traffic forecasting combining Graph Neural Networks (GNNs) and Transformer-based multi-source visual fusion. GNNs are employed in this method to capture the spatial connections between various road segments and to properly reflect the basic structure of the road network. The model's ability to effectively analyse traffic dynamics and relationships between nearby locations is enhanced by graphs representing the road layout, which also increases the outcome of traffic predictions. Recursive Feature Elimination (RFE) is employed to improve the model's feature selection process and choose the most pertinent features for traffic prediction, producing forecasts that are more effective and precise. Utilizing real-time data, the performance of the suggested strategy was assessed, enabling it to adjust to shifting traffic patterns and deliver precise projections for intelligent transportation management. The empirical outcomes show exceptional results of performance metrics for the proposed approach, achieving an amazing accuracy of 99%. The results show that the suggested technique’s findings have the ability to anticipate traffic and exhibit a superior level of reliability which supports efficient transportation management in smart cities.