Using Argo to Train Predictive Models

As a Software Engineer 2 on FlightAware’s Predictive Technology crew, Andrew Brooks continually works to maintain, analyze, and improve FlightAware’s predictive models as well as the software and infrastructure that supports them.

Foresight is the name we’ve given to FlightAware’s suite of predictive technologies, which use machine learning models to emit real-time predictions about flights. These models draw upon datasets that combine thousands of data sources and incorporate routing and weather data to forecast future events in real-time.

In this blog post, we’re going to focus on Foresight’s real-time ETA predictions. Specifically, Foresight can provide two kinds of ETAs: an estimated “on” time (when a flight lands on its destination’s runway) and an estimated “in” time (when a flight has finished taxiing to its gate at the destination). Foresight ETA predictions require two machine learning models for each destination, which provide the “on” and “in” predictions, respectively.

Training the machine learning models that power Foresight ETAs is not an easy task. In order to support thousands of destinations around the world, we need to train about 3500 different models. To make matters worse, these models must be retrained once per month, from scratch, to ensure that they’re able to adapt to any changes in real world conditions.

Leave a Comment