Difference between arm movements in butterfly and freestyle?

Hi - this is John Anthony from Swim.com. This is a great question and one that is, honestly, hard for us to answer. Here is why...  

Background
We use machine learning (ML) to automatically classify your stroke type as one of the four major stroke types (fly, back, breast, free). The power of ML is that the application to different types of swimmers is going to be much better compared to if we tried to implement a set of hard coded "rules". This is the true of many types of classification problems; not just the application of trying to identify swimmer's stroke types. However, there is a downside of using a ML approach. Debugging why something was classified as something (e.g. a stroke classified as freestyle) can be nearly impossible if not impossible. This gets more technical than you and others may care about but the issue comes down to the number of variables (aka features) that are used to drive the model. The more variables that are used, the more challenging it is to figure out why one instance / sample (in our case a stroke) is being classified as one thing vs another. The net result is that we can't say if the misclassification of your fly is due to what's happening in the recovery part of the stroke or the pull phase or a combination of characteristics across the entire stroke cycle.

The models we built for stroke type classification come from a fairly wide range of swimmers types (fitness, triathletes, competitive swimmers, etc). However, for fly, there's a kind of self referential problem in that, for the most part, swimmers who do Fly skew towards the more competitive / serious end. Meaning, there are not nearly as many fitness swimmer swimming fly compared to competitive swimmers. This means, our models for identifying fly likely skew more towards one type of swimmer compared to the other stroke types.

What we do know is that the overall "energy" we see in the motion data for these types of swimmers swimming fly is much greater than the energy we see in these same swimmers swimming freestyle. By energy, we mean the overall change in acceleration - meaning, swimming fly just looks a lot less "smooth" from a motion sensing perspective compared to freestyle. In fact, we sometimes have the opposite issue as result of this. That is, the stroke typing models will sometimes classify a very fast / aggressive freestyle as fly. This recently happened to me swimming the 50 Free TYR L1F1 race over the weekend.

Recommendation

With the background above, there are two suggestions I can make.

(1) keep practicing and get feedback from other experienced swimmers or coaches.The more you improve your technique the more likely the models will correctly identify your fly.

(2) After the swim, use the Swim.com app to edit your swim and fix the stroke type (see link below on how to do this). This will not only correct the stroke type (which can be important for Leaderboards, Personal Records and Race-based Challenges) but we can also use this correction to continue to refine our algorithms and models.

https://support.swim.com/hc/en-us/articles/360049709211-Editing-Swim-Data

So this is why I don't have one of those electronic do dads to confuse me even further!

So this is why I don't have one of those electronic do dads to confuse me even further!