UCL machine-learning forecasts of transfer and security passenger flows at Heathrow

Heathrow's Airport Operations Centre uses machine-learning forecasts developed with University College London to predict transfer passengers' connection times and the number of passengers arriving at immigration and security. According to UCL's REF 2021 impact case study, the work was led by Professor Bert De Reyck, first tested in a live trial on 19 July 2016 and implemented in 2017.

The models produce quantile forecasts, a predicted spread of likely outcomes instead of a single number, both for each transferring passenger and for aggregate flows. The case study says operations managers use them to identify passengers likely to be late for connections, alert airlines and ground staff, and adjust resourcing at immigration and security in 15-minute intervals.

UCL reports that a back-testing study found cost reductions of between 12% and 54% at immigration and security compared with Heathrow's legacy system. That is a back-tested estimate, not a measured operating result; the case study also reports that the model's accuracy beat the historical averages Heathrow previously relied on.

The evidence is the REF 2021 submission, which UCL published on its impact pages in April 2022. Heathrow has not said publicly whether the system is still used in this form.