SITA Lab has successfully demonstrated, using machine learning, that it is able to predict flight delays up to 6 hours before their expected arrival. This breakthrough will have a profound impact on providing more responsive airport operations and limit the impact on passengers.
The challenge of smoothly managing the disruption of flight delays has longed weighed on airline and airport executives’ minds. How could you stare into a crystal ball and determine which flights would be delayed and by how long?
Flight delays and disruption costs the industry an estimated $25 billion every year. The impact of this disruption was identified as one of the biggest challenges facing air transport today and an area where SITA could invest to help its members and the wider industry find a solution.
SITA Lab, SITA’s research arm, in 2017 took up the challenge to build just such a crystal ball by predicting flights delays up to 6 hours in advance. Using commonly available data and machine learning, the SITA Lab successfully demonstrated that they could make such predictions with reasonable confidence.
The team began working with a major Asian airport that were seeking a solution that would provide better insight into aircraft arrival and departures. Key challenges facing the airport were that they had limited visibility on arrival traffic and high variability of landing times due to weather and congestion. This was having a dramatic impact on the airport’s ability to effectively manage everything from allocating runway slots and gates to providing the right resources needed for aircraft turnaround and personnel at security or immigration.
“When we approached the airport, they were excited by the possibility that we could provide predictions on flight arrivals. Over a six-month period, we used various sources of information such as weather, NOTAMs (notice to airmen), flight movements and other flight data to predict six hours ahead of time the expected arrival time,” says SITA Lab’s Thierry le Gall.
“Using sophisticated algorithms, we were able to provide an accurate prediction of within 15 minutes of the flight arrival for around 80% of flights 6 hours before touch down. Building on our successes, we are improving the prediction accuracies as well as extending the predictions up to 12 and 24 hours before gate arrival.
‘Of course, these performances can vary over time and cannot be guaranteed for all airports. But the beauty of machine-learning is the more we can provide quality data and the more we learn from past predictions, the more accurate our predictions become.”
The benefits of this trial for air transport will be tremendous
“We believe that by providing more certainty to the fluid nature of flight movements and the implementation of proactive planning of the industry’s resources in anticipation of this fluidity, will be a major step forward,” says Sebastien Fabre, VP Airports at SITA.
“Today, most of our industry’s resources are planned and deployed based on carriers’ published flight schedule. With an industry-wide on-time performance of around 76%, this is an area where more accurate data could dramatically improve efficiency through proactive planning. This will directly improve the passenger experience of nearly one flight in every four.”
SITA’s work with machine-learning AI to predict with precision the anticipated arrival time of flights well in advance, allows all supporting resources to be accurately scheduled. For example, using these predictions, airports could anticipate and plan for an unexpected influx of passengers while for airlines this could help pre-empt the impact on passengers that will not make their connecting flight.
“We continue to work with various airports globally to not only further improve the accuracy of our predictions but also to understand how we can best use this information to better manage the operations of an airport,” says Fabre.
“We think the benefit of this technology will be welcomed by airlines and airports alike and we are excited about rolling out this as part of a portfolio of products that will help the entire industry better anticipate delays and manage disruption.”