0
$\begingroup$

enter image description here

Hi everyone

I have a data as on the image. I would like to derive a line that separates the data (like the red line), but not sure what algorithm can be used.

I thought about k-means in 1-d data, but then the high points (i.e. y-value ~0.01 are not really in the same cluster with the data point of 0.002).

Thanks a lot.

P/S: In fact, the values here are reconstructed MSE from h2o Autoencoder, and now I want to find an automatic way to determine the threshold, to say a point is an anomaly or not.

$\endgroup$
1
$\begingroup$

First of all, there is nothing wrong with using a threshold.

If you can't manually choose the threshold, use standard statistics. For example trim the largest 10% of points. Then compute the mean, standard deviation, and use 2*stddev as threshold.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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