0
$\begingroup$

I have to identify the different operational states of a server. I have readings related to the different sensors of the server ( like temp sensor,fan speed sensor,job load sensor etc).The data I have information about some operational states (normal, high temp , high temp and high fan speed) etc. What ML algorithms should I use to identify if any other operational state,state which the training data has not seen, comes up?

I have used several clustering algorithms. I expected Gaussian Mixture models to work well, but they fail to indicate a new operational state of it comes up.

I used LSTM and looked at the residuals, but had to look at the residuals of each parameter to identify diff states.

$\endgroup$
0
$\begingroup$

A possible way to do this is to learn some compressed latent representation of your training data, and check how well new data matches this representation.

In practice, you could try training an autoencoder on your training data, by minimizing the reconstruction error between its input (your raw features) and its output (the reconstruction).

After training, you can check whether a new operational state comes up by checking if the reconstruction error is larger than a threshold. The intuition behind this is that if the new data belongs to a known or normal operational state that the model learnt from the training data, the model will be able to reconstruct it properly and the error will be low.

| 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.