Forecasting passengers with machine learning at KLM


Nobody likes throwing good food away, while everybody likes to get a fresh meal on a long-distance flight. In to order take just the right amount of meals on board, KLM Royal Dutch Airlines successfully executed a project to improve their forecast of the number of passengers on-board every flight. In this talk, we walk you through the full-cycle process of designing, developing and industrializing a machine-learning model for this supply chain management use case. We discuss the requirements, data sources, evaluation metrics, gradient boosting decision trees algorithm and microservice architecture. Throughout the talk we highlight challenges, solutions and learnings.