0
$\begingroup$

I read that neural networks, SVM and neuro-fuzzy don't support "mixed variables."

So what are those exactly? Does it refer to mixed types (numeric and non-numeric)?

And if so, does that mean the input has to be all numeric? what about the output?

$\endgroup$
  • $\begingroup$ This would be easier to answer if you included the source and context you read it in. $\endgroup$ – Sean Easter Apr 6 '17 at 1:12
  • $\begingroup$ @SeanEaster this is the source ieeexplore.ieee.org/document/7451491 $\endgroup$ – engineering student Apr 6 '17 at 2:41
  • $\begingroup$ @SeanEaster and there isn't much "context" there is just this comparative table that states each classifier's advantages and disadvantages and it says "does not support mixed variables" for those algorithms I mentioned before and that is it $\endgroup$ – engineering student Apr 6 '17 at 2:42
  • $\begingroup$ The link isn't all that helpful as the paper is gated. In general, "mixed variables" refers, as you indicated, to information of mixed types or scale. And again in general, when there is a limiting reference in the literature to "not supporting mixed variables," this is an indication that continuously distributed information is required, i.e., not even 0, 1 dummy variables are to be used. This is a limitation of many multivariate methods, e.g., regression (use ANOVA when the predictors are categorical), PCA, factor analysis, k-means clustering, etc. Use finite mixture models instead. $\endgroup$ – Mike Hunter Apr 6 '17 at 13:31
0
$\begingroup$

From the paper:

First step: understanding what type of data is available. Educational data is usually in the form of either numeric or categorical.

So it would seem the authors intended "mixed variables" to mean datasets entailing both numeric and categorical data.

Though, it's worth mentioning that categorical data can be transformed to numeric via one-hot encoding. On this site, related questions abound on this very topic, and this example from one (prominent) open source SVM implementation uses it to predict flower category. It's not really clear why the authors presented these methods as not supporting categorical data. (Perhaps the implementations they had available at the time did not support transformation.)

$\endgroup$

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.