Optimising Airline Food: Using Machine Learning to Forecast Demand
A complex and interesting project that I recently led was to develop and implement a machine learning model to forecast demand for perishable products on-board aircraft. In other words, the project was to optimise loading of fresh food items on flights in order to minimise waste and minimise sell outs. This in turn was aiming to maximise on-board margin whilst keeping customers and crew happy.
The project was very challenging as it involved stakeholders in the supply chain, the in-flight retail team, software suppliers (in-flight selling software), and of course the cabin crew. I encountered many data challenges in the early phases of the project, and implemented several improvements to data capture technologies (in conjunction with the iPad software developers), as well as thorough and robust data cleaning and quality reporting.
With better data (although still not perfect!) I was able to develop a machine learning model using a combination of technologies (Alteryx, AWS tools, and R). The model predicted for future flights (a week out) the expected demand for perishable products, and applied uplift logic based on risk calculations to recommended a loading level.
Following a successful trial, the model was rolled out across the airline, and delivered a 22% increase in margin per passenger within the fresh food product category.
Read details of the project here.
