0
$\begingroup$

Suppose I have a feature "Pool Size" whose levels are: Big, Medium, Small.

Surely if a house doesn't have a pool it certainly won't have a value for "Pool Size".

Q1-For this particular instance/house, should I consider the value for "Pool Size" missing ('NA') or should I makeup a new category for 'Pool Size' dedicated to houses that do not have a pool size lets call it 'None' or 'NoSize'?

Q2-If I choose to makeup a new category for 'Pool Size' then my feature would become categorical instead of ordinal. Wouldn't that be a problem as I am essentially changing the nature of the variable?

$\endgroup$
1
$\begingroup$

If you are able to get this information, it would be good to include it into the dataset. Your proposal to add a value none to the size feature sounds okay, after all, no pool is still smaller than a small pool. However, if you want to keep the feature ordinal as it is, you could add a new binary feature has pool to indicate this information.

$\endgroup$
1
  • $\begingroup$ When you said "no pool is still smaller than a small pool" this helped me understand that it's ok to let 0 values if pool size was a numerical feature or a value of 'None' if pool size was a categorical feature (as presented in my example). This makes sense. However, I don't think your strategy of adding a flag feature 'has pool' would help in keeping the feature ordinal. At least, I don't see how. Perhaps you could expand on that further $\endgroup$ – Imlerith Feb 3 '18 at 3:13

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.