I have users' reports about an accident. I want to know how to make sure that the number of reports is big enough to take that accident as a true accident and not spam.

My idea is to consider a minimum number of reports in a specific time interval, for example 4 reports in 20 minutes are good enough to believe the existence of that accident.

My question is how can I choose the minimum number of reports and that time interval? Is there some logic to make that decision?

  • 3
    $\begingroup$ This kind of question,should fit here CrossValidated $\endgroup$
    – quintumnia
    Feb 10, 2017 at 9:19
  • 2
    $\begingroup$ Your problem is an outlier / unusual event detection problem. Countless methods exist for this. A threshold is very simple but, in my opinion doesn't reflect the reality. What if you have an accident alone, at night, on a small countryside road? These situations exist and discarding them per se is doesn't sound like a good approach to me... I agree Cross Validated SE would be more adapted. $\endgroup$
    – Eskapp
    Feb 21, 2017 at 20:09

2 Answers 2


If the only feature you're classifying on is the number of users making a given report, then this isn't really much to do with AI/ML. Just pick a number based on your subjective judgment and go with it.

OTOH, if you can include details of the report itself (as well as the number of reporters), I think you might be able to build a bayesian classifier that would be useful. If you could consider location, weather, time of day, number of reporters, etc., it seems like you might be able to get something useful put together.


It's a trust-level problem, so your judgment is the best to decide what would be rhe threshold.

You can help your decision making by trying and visualize how many accident (in %) you've left out... this can be an indicator of good threshold. You don't want to throw too many of them. But only you know what is good and bad in this case


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .