I have a response variable series which will be generated randomly in a fixed interval [0-100] base on every second, and I want to detect the event when the new generated data is significantly greater than data of last second, and send alarm message to me.

So, I calculate the difference of response variable by 1 lag and divided by difference of time (slope), than use bootstrapping to construct the 95% confidence interval of response's 90% percentile, if the new data is greater than Upper Limit, I define it as unusual.

In fact, the quantity of data I have is extremely large, and these data update frequently. So, it will be horrible to resampling and calculate the CI, even it could be a way to solve problem. And it's looks like not necessarily to sampling because I have large sample.

Are there any better way to do? Thanks!


1 Answer 1


One option is to collect the empirical distribution of 1 lag (i.e., how much change happens over one time step?). Then set the threshold based on a percentage of this empirical distribution.

This is often done with latency. An unusual latency event is defined at 95% or 99% percentile.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.