Performance Predictions

What are Performance Predictions

Performance Predictions gives subscribers estimated completion times for key running race distances based on their historical Strava activity data. The race distances supported are the 5K, 10K, Half, and Full marathon. Performance Predictions do not consider any terrain or altitude variability for the race and assume that an athlete runs the race on a flat course, similar to a track. Predictions are only available to subscribers and can be found on the Progress section of the You tab.

How they work

To see predictions, a subscriber must upload at least 20 run activities within a rolling 24-week (about 5 and a half months) window. This threshold ensures that the machine learning model powering the feature has sufficient data to make a high-quality and accurate prediction. The model generates a new set of predictions for the subscriber after each run upload and after three days without any run uploads. Subscribers who have not uploaded enough run activities within the rolling window will see a cached set of predictions from the last time they had enough uploads. The predictions will update once a subscriber resumes uploading and hits the activities threshold.

Our methodology

Strava’s Performance Prediction feature is powered by an ML model that leverages over 100 athlete data attributes, including all-time run history and top performances. Unlike other race predictors that rely on theoretical inputs like estimated VO2 max, Strava only uses real activity data to predict race results. The system also leverages the performances of athletes with similar training histories, so estimated times are realistic and based on what has been achieved by other users with similar capabilities. 

Times for each race distance are calculated independently, which leads to greater precision. For example, an athlete training for a marathon – running more weekly volume and focusing on longer intervals – may see significant improvement in their half-marathon and marathon predictions but not see equivalent improvement in their predictions for the shorter distances. Similarly, an athlete focused on shorter distances – emphasizing speed and power in their training – may see more improvement in their 5K and 10K predictions than they do in the longer distances where those capabilities are less important.