Starting Date: June 2024
Prerequisites: Having good knowledge of Python programming is essential. Having some basic knowledge of Machine Learning is beneficial (but not mandatory)
Will results be assigned to University: No
Last year, only 67% of UK trains arrived on time – the worst performance in over a decade. That was not just an inconvenience; it was 8.3 million compensation claims, over £150 million drained from the system.
The problem with existing AI systems is that they treat each train in isolation and provide a single best-guess. This is highly reckless as rail operators are rerouting thousands of passengers in real time.
Therefore, this project will develop a machine learning system trained on millions of UK train records over four years. The model aims to capture how delays propagate across the entire rail network. More crucially, it will provide a statistically valid confidence guarantee for every prediction (i.e. not just “your train will be seven minutes late,” but “seven minutes late, with 90% confidence”).
This project is supported by a UK rail operator, providing direct industry validation and a pathway to real-world impact.
Students are welcome to email (Khuong.Nguyen@rhul.ac.uk) for informal discussions.
Readings:
- Feng Xu, Khuong An Nguyen, and Zhiyuan Luo. “Reliable Train Delay Forecasting with Conformal Prediction”. The Importance of Being Learnable. Cham: Springer Nature Switzerland, 2026. 387-410..
- Rui Luo, Xiaoyi Su, Khuong An Nguyen. “Uncertainty quantification in train delay prediction with Conformal Prediction”. Philosophical Transactions A, Royal Society.