Prediction of Flight Delays

Challenge

Creating and tracking a flight schedule is an extremely complex and multifactorial problem. Whether airline passengers arrive on time at their destination airport each day depends on many factors, most of which are beyond direct control. Our client, an international airline, aimed to improve existing delay forecasts using modern machine learning methods.

Approach

Together with our client, we developed machine learning models at various hierarchical levels to support network planning. The models process a multitude of external factors, such as weather forecasts and capacity planning at international airports. The hierarchy of the models considers different levels of information, with departure and destination airports determining the information level and thus the model. We developed a framework to generate automated daily forecasts for all client flights in the coming days.

Result

Despite the complex dependencies within the flight network and its ecosystem, we helped our client incorporate predicted delays into the current flight schedule through the targeted use of modern machine learning models. This provides the operations team with a tool that encourages proactive action and helps ensure passengers reach their destination on time.