Machine Learning: Runalyze can now guess your activity type

, Posted in Features

The feature presented here is a beta feature that is only available for Supporter and Early Premium users. You can sign up for a supporter or premium subscription via your upgrade page.

All the analysis that Runalyze offers you for your training data is usually based on the simple but important assumption that you record your training data completely and correctly. This includes not only the synchronization of the training file of your watch, but also the entry of the other “meta” information: Activity Type, Equipment, Tags, Route, Title, Notes, RPE, Subjective Feeling, and so on. While much of this may only be important when looking at details, the training type is a very crucial information to correctly classify the training in the weekly overview. The good thing is: Runalyze can try to guess the training type of your activities in the future.

The beta feature for guessing the training type must be activated manually in the configuration.

This feature is still a beta feature and optional. You have to enable it in your settings first. Runalyze then analyzes your previous assignment of activity types and uses machine learning to try to identify features in your training data that can be used to automatically determine the activity type. Long runs are long, recovery runs are short and relaxed, tempo workouts have a high proportion in the high heart rate zones, and competitions are almost exclusively high intensity.

So far, so simple. However, since each athlete can define and assign the activity types individually, and of course the question of what is “long” or what is “intensive” is answered differently for each individual, these models must be fitted individually for all athletes. Therefore, in order to “train” this ML model, sufficient workouts must already be available and assigned to the correct activity type. If, on the other hand, every workout has been marked as an endurance run so far, every workout will automatically be recognized as an endurance run in the future as well.

Of course, the feature is not limited to running, but is applied to all sports. However, the whole thing naturally only makes sense if there are also different activity types and these can be sufficiently clearly distinguished by the recorded training data (primarily duration, distance, power, variability and heart rate zone distribution). For sports other than running, it can therefore often happen that the feature is not noticeable.

Of course, guessing the type of training is not always easy, even in running. The differences between an endurance run and a recovery run, or even the difference between fartlek and interval training, can be fluid. In addition, depending on the training condition, a 15-kilometer run can of course be considered a “long run” at times, but in marathon preparation perhaps only from 30 kilometers.

In our tests, the models mostly achieved an accuracy between 65% and 85%. Even in the worst case of 65%, Runalyze would still detect the correct training type on its own for 2 out of 3 workouts. – This is significantly better than before, because up to now the “default training type” was simply used for all trainings.

You have questions, comments or wishes about this or another feature? Let us know!

Start your analysis now

Create a free account and start analyzing your activities.

You can also sign up with your Facebook or Google account.


Any questions left? We are there for you. Ask your questions in our Forum or Facebook Community.

Need help?

Do you have troubles with your account?
Contact our Support.