0
$\begingroup$

I would like a direction for a ML technique to predict when an anomaly will occur in a system, like when the CPU use differs from a normal state.

My data consists of system metrics collected from AWS EC2 instances such as CPU Utilization, Disk operations, Memory Use etc. Each row of the data consists of a timestamp and values for the metrics.

My idea was to perhaps build a LSTM Recurrent Neural Network since I´m working with time series. Do you think this would be a good approach for this problem?

$\endgroup$
0
$\begingroup$

You can also perform ARIMA decomposition with regressors. ARIMA is the model in which each step of the time series depends on the previous steps of it and on the previous error terms. Regressors are your additional features, which could help determine next value of the time series.

PROS: Easy to implement, can auto-tune proper ARIMA parameters.

CONS: Linear dependence on regressors you will provide, therefore may underfit.

$\endgroup$

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.