3
$\begingroup$

I need to collect several large datasets (thousands of samples, dozens of features) for regression with only categorical inputs. I already look for such datasets in the UCI repository, but I did not find any suitable one.

Does anybody know of any such dataset, or of any additional dataset repository on the Internet?

$\endgroup$
  • $\begingroup$ Since you can create categorical variables out of numeric ones, maybe you should drop "categorical" from the title. $\endgroup$ – JenSCDC Nov 12 '14 at 11:22
  • $\begingroup$ Creating categorical variables out of numeric ones is always a possibility, but I'd rather find specific categorical datasets. $\endgroup$ – Pablo Suau Nov 12 '14 at 11:36
  • $\begingroup$ Why do you want that? $\endgroup$ – JenSCDC Nov 12 '14 at 11:39
  • $\begingroup$ I don't want to introduce any assumption about the structure of the data. I have the impression that creating categorical features from numerical ones would have that effect. But I may be wrong. $\endgroup$ – Pablo Suau Nov 12 '14 at 15:32
  • $\begingroup$ I don't see why it would. In a regression, a variable indicating an income of < 50k is no different from one indicating an income of >= 50k. $\endgroup$ – JenSCDC Nov 12 '14 at 17:05
2
$\begingroup$

I would like to recommend to check the following open data repositories and meta-repositories (they are not focused on categorical data, but I'm sure that many data sets, listed there, contain such data):

Also check built-in data sets in the open source software Parallel Sets, which is focused on the categorical data visualization: https://eagereyes.org/parallel-sets.

$\endgroup$
0
$\begingroup$

Try the 1998 KDD Cup dataset. Its a regression problem with categorical and integer predictors. For your task, you could either treat integer predictors as categorical or ignore them completely.

$\endgroup$
0
$\begingroup$

All you need are data sets with enough records and enough features for your purposes. You can simply convert any continuous variables into categorical ones by grouping. Some sources for large sets can be found by a search for "large free data sets". If you are dead set on lots of categorical data, try insurance data (given that I'm an actuary, I should have thought of that earlier). Those tend to be laden with categorical variables, as I well know from first person experience.

$\endgroup$

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.