A team of engineers from the University of Illinois-Urbana Champaign built an artificial intelligence (AI) model to improve predictions about transportation delays using British Railway data.

Using graph neural networks — a neural network operating on the graph structure that is used for node classification wherein nodes are connected via edges — the team built the Spatial Temporal Graph Convolutional Network model.

This model, according to its developers, makes predictions about delays along portions of the British rail network, often up to 60 minutes into the future.

Although AI capable of making such transportation-related predictions already exists, that data only concentrates on the full trip, from start to finish, and not on the delays encountered along the way. The new system, however, can be used to approximate what portion of the trip will experience delays.

Going forward, the team wants to further develop the AI models to better explain why a delay is happening where and when it is — something that is not currently understood via AI models. This improved understanding, according to researchers, might improve the response to such delays and, ultimately, transportation outcomes.

The study, "Railway Delay Prediction with Spatial-Temporal Graph Convolutional Networks,” was presented at the IEEE 2020 International Conference on Intelligent Transportation Systems.

For more on the presentation, watch the accompanying video that appears courtesy of the University of Illinois.