Exploring the Efficacy of Graph Neural Networks in Predicting Traffic Flow in India

Accurate and real-time traffic forecasting is essen- tial for managing vehicle movement, reducing congestion, and generating optimal routes. However, existing traffic prediction systems rely on datasets specific to certain regions, making them unreliable for Indian traffic and road networks due to the lack of suitable training data. To address this issue, in this paper, we propose a Dataset for Indian Vehicular Traffic Analysis (DIVA) capturing the traffic features of the Indian Province. By training Graph Neural Networks on this dataset, we achieved comparable results to the state-of-the-art datasets (PEMSD7, PEMSD8), demonstrating the efficacy of the DIVA dataset in accurate flow prediction. Using this dataset, appreciable accuracy was achieved for traffic speed prediction using GNNs with STGCN achieving the lowest RMSE values. Moreover, we propose a novel route navigation approach that utilizes predicted future traffic speeds at nodes, represented as a future graph thus, provide a path that maximizes the traffic speed hence, minimizing the travel time.