3
$\begingroup$

I plant engineer asked me to analyze some data they have regarding production. It has multiple materials quantities, machines parameters, and final product quality (which he wants to predict). The data has a time stamps every 3 hours for the last year, 25 parameters (all continuous variables). I tried to detrend the series but a colleague says this could backfire - I didn't really understand why.

I am new in this position and have a lot to learn so any advice will be really appreciated.

$\endgroup$
3
  • $\begingroup$ Do you have a reason to suspect there is a trend in the final product quality independent from your input variables? If one of your inputs has a trend, you may have a problem separating input effects and trends. $\endgroup$
    – MSalters
    Jan 31, 2018 at 14:16
  • $\begingroup$ You mean there may be no reason to detrend or treat the series? What do you mean by input effects? Thanks! $\endgroup$
    – Luis
    Jan 31, 2018 at 19:04
  • $\begingroup$ Say that there is a trend in both your input (machine parameter, wear&tear?) and in your output (product quality dropping?). Detrending both might affect the quality of the correlation between them. $\endgroup$
    – MSalters
    Jan 31, 2018 at 22:21

1 Answer 1

5
$\begingroup$

You don't need to always de-trend time series data before analyzing it. Detrending is required for some types of time series models, such as ARMA/ARIMA models. For other models, you don't need to detrend the time series.

If you do detrend, the series, then you need to make sure to "retrend" by applying the reverse transform after you've generated your model.

In your case, you seem to using multiple variables, that might be what your colleague is referring to: detrending is used mostly when regressing a variable against itself - that is using past values of a variable to predict future values of the same variable. But in the case of multiple variables, deterending might remove correlations between different variables.

$\endgroup$

Your Answer

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

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