1
$\begingroup$

I have to establish an ML-based model where I predict precipitation in a complex terrain using multi-year daily observations from 50 stations. Besides a dozen of continuous variables, predictors include three variables that reflect topography: elevation, slope, and aspect. As these three variables do not change for a single station, I have doubts that the model will count on these during the training (I haven't yet started the analysis, still compiling the data frame).

  1. Are my concerns valid?

I'm thinking about writing a function that will randomly alter these three static variables per each observation in a data frame by a small margin, e.g +-2%.

  1. Would there be major caveats behind such an approach?
$\endgroup$

1 Answer 1

1
$\begingroup$

You have 3 variables that refer to a particular station.

In your training set, you have only one station? -- If yes then the best is to drop them

In your training set, you have more? -- Then they can have distinct values so leave them.

If your training set has one station, and then in your test set you have another station -- Drop them. Your model wont be able to learn from them. But there are high chances that you have a data set shift in your model and it wont perform well.

$\endgroup$
7
  • $\begingroup$ Thaks @Carlos. I have 50 stations, but for each station those three variables do not change. E.g. the combination of the three variables for any single stays the same across all observations $\endgroup$
    – tabumis
    Oct 26, 2021 at 9:19
  • $\begingroup$ then its okey! Keep them. How many datapoints per station? $\endgroup$ Oct 26, 2021 at 9:29
  • $\begingroup$ I will reccomend you to do split your data within stations, 30 for training, 10 for validation and 10 for test $\endgroup$ Oct 26, 2021 at 9:29
  • $\begingroup$ You are more than wellcome, dont forget to (i) upvote if you liked (ii) mark as accepted if you think that is the answer to the question $\endgroup$ Oct 26, 2021 at 11:15
  • $\begingroup$ So ML methods will somehow learn from these static predictors? E.g. regression model would discard those static predictors because they are not varying. Yes, Im planning to set aside some independent observations for validation. Thanks for your inputs! $\endgroup$
    – tabumis
    Oct 26, 2021 at 16:37

Your Answer

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

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