Automatic flagging suspicious activities and excluding from segment data

[steven_willems]

For a lot of activities it's pretty straightforward to detect when an activity is suspicious and probably should be flagged. Often it's not even intentional, just forgetting to stop the recording and get in the car can be enough to have a malicious activity. Including those activities in the segment boards is pretty annoying and could be prevented. Rather easily, actually. if a human can see it in an eye blink, a machine should be able too these days.

That being said, how could such a feature be implemented?

• for each activity type, define some static rules: e.g. max speed for (e-)bike over certain value, or clearly two different AVG speed can be detected on segments that are not downhill.. And those are just simple suggestions. I bet based on the data available to Strava, if you do some big data analysis, I bet one could define some very accurate rules. 
• Apply these rules to the recorded activity and mark it suspicious if needed
• If suspicious, do not include the recording into segment stats
• If suspicious, notify the user that the activity type might be wrong, or that the activity probably wasn't stopped.. It could even be suggested to adapt the recording with a prepared shortened activity.
• Allow the possibility to the owner of the activity to confirm the activity was correct. if so, this could be reported to Strava and manually be validated by Strava. Just like now other people have to report wrong data. It would make the flow inverse and much easier to follow up (if the rules defined above are well defined).
• when the activity is changed, re-evaluate the activity.

There's a lot of benefit to such an approach. First of all, we the users won't be annoyed by all these wrongly recorded activities. We love Strava. We love Strava segments, but it's to not much use if the data in the segments is bogus. Reporting can be avoided. Turn the system around. Keep the good users happy 🙂
Secondly, for Strava itself, having quality data is key I would assume. So get the data filtered when it's entered. Data cleansing can be hard, as a software engineer I'm well aware of this, but when done good and early, it can be very, very rewarding.

[Jane]

Hello @steven_willems 

Thanks for posting about this.  Segment leaderboards are an important part of Strava and we appreciate you taking the time to share your thoughts.

We do have an existing functionality that runs on the backend and 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 do prevent a large number of invalid segment times from ever appearing on leaderboards.

We are constantly working on improving our algorithms for this process and appreciate your feedback on this important subject. Our longterm goals are to incorporate machine learning into this mechanism.

If you notice invalid times on segments, the first step is to use the activity flag tool.  If the Athlete doesn't correct their behavior after getting their activities flagged, you have the option to report that Athlete to Strava Support so we can follow up.

I would encourage you to check out this existing discussion on the topic here.  Again, we appreciate you taking the time to share your thoughts.

[anchskier]

@Jane - As we have mentioned multiple times before, your automatic flagging is terrible.  You claim it compares runs to the 400m world record time, yet I constantly see runs that show times way faster than that on leaderboards.  Even if that worked, it really needs to be a graduated scale.  The 400m world record time would be good for runs up to maybe a mile.  When you have runs of 3+ miles, they should be compared to the world record 5k time.  Runs over 6 miles should be compared to the current 10k world record.  This would filter out far more incorrect files than the current system does.  

[steven_willems]

@Jane Hi Jane,

thank you for your quick reply. Good to read there already are systems in place to detect bad recordings. However, fact that I didn't notice them and thought they were none existent might indicate that they are simply not good enough, at all. I've been flagging the previous days multiple activities that were so obvious a blind could have spotted them. I still have to see an e-bike that goes over 75km/h on a non descending road. As far as I know, that would be even illegal in most countries.

So, Strava should step up it's game. As I stated, the data quality should be Strava's priority number one. In the interest for itself, but also for its users. When the basic functionality, i.e. data can't be trusted, than what's the use?

And while I'm at it, is there even a product manager at Strava? The experience in the webapp vs mobile could be so much better, so much better in line.. what are you guys doing over there?

[Jane]

Hello @steven_willems 

Thanks for your feedback, we do appreciate it.  (And you also @anchskier)

I would definitely recommend posting your comments and feedback in this existing discussion in order to give them the most visibility. 

In regards to Mobile vs. Web, we have a lively discussion on that here, and we would welcome your contributions and thoughts.  Thanks again.