Researchers have developed a new model called RSTGCN to predict train delays at railway stations. The model uses graph convolutional networks and incorporates features like spatial attention to improve predictive performance. To test the model, the researchers created a comprehensive dataset for India's entire railway network, which includes over 4,700 stations across 17 zones. They compared their model to several state-of-the-art baselines and found that RSTGCN outperformed them in terms of accuracy, reducing mean absolute error by 18%, mean absolute percentage error by 14%, and root mean squared error by up to 8%. --- Why it matters: This matters because accurate train delay prediction can help optimize railway operations and reduce congestion. Engineers working on AI for transportation will be interested in the RSTGCN model's ability to handle large-scale datasets and improve predictive performance. Source: https://arxiv.org/abs/2510.01262