Dynamic thresholding for motors using in anomaly detection with autoencoder ( PHM related problem) | Coursera Community
Coursera Header
Solved

Dynamic thresholding for motors using in anomaly detection with autoencoder ( PHM related problem)

  • 28 October 2019
  • 1 reply
  • 63 views

Badge +1

When I use autoencoder for anomaly detection, I need to set the threshold. Is there any way to set the threshold automatically? Especially for motor maintenance, if there are a large number of motors, how to set the threshold value quickly to judge whether it is good or not?

icon

Best answer by hamster 3 November 2019, 10:47

The question is a bit vague. But here are a few pointers.

You might want to reconsider the model you are using. For example, some models ouput a class label. Other models (logistic regression) will output the probability (p) that the element is label X. 

If you have a probability, then you can set a threshold based on that. “i.e. show me all the items where I am 70+% confident that the item belongs to class X”. If you want to threshold the data in some other way, then you will need to be more specific in your question.

 

View original

1 reply

Userlevel 5
Badge +4

The question is a bit vague. But here are a few pointers.

You might want to reconsider the model you are using. For example, some models ouput a class label. Other models (logistic regression) will output the probability (p) that the element is label X. 

If you have a probability, then you can set a threshold based on that. “i.e. show me all the items where I am 70+% confident that the item belongs to class X”. If you want to threshold the data in some other way, then you will need to be more specific in your question.

 

Reply