Traffic Prediction with Temporal GNNs

Forecasting with connected sensors

This 2023 Machine Learning with Graphs course project explored traffic prediction with a spatiotemporal graph neural network. The problem combines the temporal history of a sensor with relationships between sensors, rather than treating each time series as completely independent.

I contributed as one of four authors, alongside Marko Ivanovski, David Ocepek, and Matej Miočic. The public repository contains the implementation and technical report. It supports the team attribution; it does not establish an exclusive component-level division of work, so I do not claim sole ownership of the model or evaluation.

Inspecting the work

The source and report are the evidence for the experimental setup and comparisons. This portfolio records a delivered research implementation, without promoting an unverified benchmark win or real-world traffic-management deployment. Reproducing the reported experiments requires the dataset, preprocessing, and environment described in the project.