A simple GPS data quality-check to improve leaderboard relevancy

[BrunoW]

Many segments'leaderboards are rendered useless by erroneous GPS data. 
Figures for the segment are often incoherent, with max speed being inferior to average speed.

A simple solution to address this : checking if the average speed is inferior to max speed for any performance ranking in the top 10 on a segment before ranking an activity.

This basic operation to implement and would look like this : 
if max speed < average speed, then do not rank the performance on the leaderboard.
It would only trigger if the performance is good enough to get into the top 10 which would greatly help contain the cost of such an operation for Strava all the while immensely improving the experience around segment leaderboards.

A quick and easy win !

Given leaderboards are a central feature of Strava and Strava subscriptions, especially for its most avid users (and promoters), this would pay for itself through additionnal paid users.

Additionnally, KOM/QOM on these segments would not get notified so often with false new KOM/QOM and would not have to report the activities manually every time, improving user experience and thus user retention among this group of very active users.

[Jane]

Hello @BrunoW 

Thanks so much for posting about this.

We do have a functionality that runs on the backend called "Automatic Flagging aka Themis" that effectively checks activities for anomalies. This depends on the Sport Type, but runs for instance are automatically flagged when a portion of the activity exceeds the current 400m World Record. For rides, it depends on the speed combined with a certain value for the grade of the hill. We have algorithms in place for this --  admittedly, they aren't perfect and don't catch every erroneous activity, but we are constantly improving them and appreciate your feedback on this important subject. Our longterm goals are to incorporate machine learning into this mechanism.

[anchskier]

@Jane - Here is just one of an endless number of activities that seem to continually elude the Themis automatic flagging.  They show up every single day in the running challenge leaderboards.  Who knows how many other times they show up in other profiles that don't subscribe to the challenges.  Evening Run | Run | Strava  Someone running a 1:33/mile pace for over 17 miles is just a tad bit faster than the 400m world record pace.  I'm really hoping that the system can be improved to address these files.  It would go a long way to improving the leaderboards as well as improving the accuracy of the total stats on Strava.  

[BrunoW]

@Jane 
Thank you for your response but I don't think it really answers my proposal. 

Themis seems to be a complex algorithm taking many factors into account. 
Although I am glad it exists, this is not what I am proposing. 

Detecting most problematic rides and performances is way simpler than what you describe. 
Almost all erroneous activities share one characteristic : average speed > max speed. 
Since this is mathematically impossible, excluding these activities from leaderboards through a simple rule on upload would solve the issue.

I'd be interested to know how such an incoherence can occur based on the same GPS data but that's not really my point.

Themis sounds like a great piece of tech but it seems to fail for now.
I assume Themis runs when server load is lower due to cost so why not put a basic set of checks first ?
Let's call it Gaïa to stay coherent with the greek godess theme. 
The first one being : if average speed > max speed then don't rank the activity for the affected segments.

Pretty sure a good data scientist can identify commonalities between flagged segments and add to that list of rules. Three digits average speed on flat cycling segments comes top of mind (even in metric that'd be over 4000 watts). 

This would greatly improve the user experience for little cost and effort.