I Think We'll Go to Boston - Marathon Performance Prediction

Download the
Full Paper Here
Authors

Varun Pemmaraju, Strava
Dave Hoch, Strava

Abstract

Predicting performance in response to training is the holy grail of sports science. This question captivated the Cold War era physiologists who developed periodization; was practically applied by a doctor who redefined what was humanly possible1; and was studied by Nike as they pursued Breaking 2 hours in a marathon. It is this endeavor - the marathon - that is especially of interest. Unforgiving and addicting, marathons have attracted 20 million people worldwide to the start line in the last decade. No one has studied this question in a way that takes advantage of this enormous population(not even FiveThirtyEight!). We set out to:

1. Predict marathon performance:
         a. Across a broader population than studied before
         b. Using a volume and variety of data that has not previously existed
2. Determine the most important features and use those to prescribe training personalized to an athlete’s background and goals