Follow Datanami:
May 14, 2020

For American Airlines, Machine Learning Solves an Air Cargo Conundrum

Air cargo shipments operate on a much tighter timeframe than passenger flights, with bookings typically occurring around ten days before departure. Many of these planned shipments, however, still don’t arrive at the warehouse, causing pandemonium for the air cargo carriers. In a blog post, Nvidia highlighted how American Airlines used data science to combat the chaos of air cargo uncertainty.

For American Airlines, the most challenging variable to predict for air cargo shipping optimization is whether a cargo shipment will arrive on time for departure. Worse yet, no-shows scramble the airline’s carefully planned cargo hold layouts, reducing the aircraft’s fuel efficiency, and the airline typically learns of no-shows without enough time to plan to carry other cargo instead. 

“No-shows cost us millions in lost revenue, and many times they can result in us needlessly turning away other critical shipments when we could have otherwise carried them,” said Chris Isaac, managing director of American Airlines Cargo Revenue Management. “Being able to firm up a flight’s bookings in advance allows us to recapture space that will go unused and provide it to others who need it.”

American Airlines decided to create a machine learning model that analyzes each customer’s booking to predict the likelihood of a no-show shipment. The model was trained with a year’s worth of cargo data – half a million records, each with around 20 variables – using an open-source, GPU-accelerated ML package called H2O4GPU. 

The training process was conducted on Z by HP data science workstations powered by Nvidia’s Quadro GPUs, which provided a tenfold speedup compared to CPUs. According to Tassio Carvalho, head of the Center for Machine Learning and Artificial Intelligence at American Airlines, the model now shows a prediction accuracy of “at least 90 percent.”

The airline will be running the model three days before each cargo flight. The model will flag bookings that it thinks show high risks of no-show shipments, and American Airlines staff will reach out to those customers to confirm the shipments – a much more manageable scenario than calling every customer. American Airlines is also allowing free cancellations with 48 hours’ notice to incentivize no-shows to reschedule.

“The ability to use advanced analytics to solve one of our industry’s biggest problems is a game changer for American Airlines,” said Isaac. “We have the best data science team in the industry, and we couldn’t be more excited to integrate the model into our business process.”