August 20, 2019

More data, less stress: the future of air travel

Running through airport

TL:DR;

  • When machine learning is leveraged appropriately, even problems like stressful airport connections can be eased for the good of the public.

Air travel is stressful. Long lines at security, cramped seats in coach, lost baggage, and tight connecting flights—there is no shortage of pain points in any journey. Machine learning can’t give you more leg room, but it is poised to revolutionize your travel experience.

In her recent collaborative work with Heathrow Airport and researchers at University College London, Yael Grushka-Cockayne, visiting associate professor at Harvard Business School, demonstrated how machine learning can successfully intervene to address one of these pain points—tight connecting flights.

You might also like...

As Grushka-Cockayne explains, there was enthusiasm among key stakeholders at Heathrow to upgrade existing data systems at the airport, and a consensus about the opportunity to better leverage data to improve the experience of connecting passengers—who account for roughly one-third of all travelers who pass through Heathrow annually. The question was how to do this. “People want to use machine learning and big data—all of these buzz words,” says Grushka-Cockayne, “but if they don’t know how to focus in on a very specific task that can generate predictions, it is difficult to use the technology to actually improve decision-making.”

“To build a model that could both identify connecting passengers and accurately predict their transfer time, it was essential to collate multiple levels of data…”

Grushka-Cockayne and her team spent several months working with partners at Heathrow to define the scope of their research—the development of a machine learning model that could predict a passenger’s journey through Heathrow in route to his or her connecting flight. The goal was to be able to anticipate the number of people passing through immigration in real time (enabling more efficient staff allocation at immigration lines), and also to predict whether a passenger would be late for his or her flight (allowing the airport to proactively offer supporting services).

But it wasn’t easy to capture the complexity of a passenger’s journey through an airport in a statistical model. “Most of the data was there,” says Grushka-Cockayne, “but there were a lot of very old, fragmented systems” that weren’t communicating effectively with each other. To build a model that could both identify connecting passengers and accurately predict their transfer time, it was essential to collate multiple levels of data: from the airport, from airlines, from passengers, and—most importantly—from baggage tags. As she explains, “the baggage database ended up being really useful, because bags are tagged with barcodes to the final destination. You know who started out where and where they’re going to end up.”

Once Grushka-Cockayne and her team had identified the data set, the final task was to select an analytically rigorous computing system that was still accessible to airport control staff. As she explains, “we tried to find a balance—a tool that could generate better, more consistent predictions, but that was not overly complex,” to enable seamless implementation.

“We tried to find a balance—a tool that could generate better, more consistent predictions, but that was not overly complex.”

Two years after a live trial in 2017, the results are impressive. The mean absolute error of passenger flow predictions has dropped over twenty percent after the adoption of the new model. But perhaps more tellingly, Heathrow stakeholders have since gone above and beyond the initial scope of the model to adapt the predictive technology to enhance other airport processes. “I was shocked, I was speechless,” says Grushka-Cockayne, “by how rigorously they pursued the implementation of the model. I think it is indicative of the fact that the entire attitude towards data science and machine learning applications has changed.” And this enthusiasm is not limited to Heathrow. Since the publication of their research, Grushka-Cockayne and her team have received requests from airport representatives worldwide—from Paris, Singapore, Los Angeles, and more—looking to implement similar systems at their own facilities.

This kind of model is just the beginning. The sheer amount of data being collected within the scope of air travel means that there are countless applications at the intersection of machine learning and travel. “The notion of connecting through airports will become better,” says Grushka-Cockayne. Part of the reason will be because systems improve—already, fewer people are missing flights and less baggage is getting lost. But the other equally important piece of mitigating travel-related stress is simply having more transparency surrounding the entire process. “Data has a lot to do with that,” she says. “When you miss a flight but know that you’re already rebooked, or if you can see where exactly your bag got lost along the way, it helps to remove some of the stress. Data can give us a lot of comfort these days. So the next step is just to use the data more effectively to prevent these kinds of things from happening.”

Related Topics

People in this Article

Keep Reading

Next Up

Next Edition | October 7, 2019