1
$\begingroup$

I would appreciate your input on which predictive ML model(s) could fit our dataset the best.

The primary features of the dataset are x,y (continous) and c (factor with around L=500 unordered levels). The secondary features (~20) are both continous and factors. We know that the lowest-order correlation is x~y. We also know that modeling x~y for each level in c improves the fit significantly.

We could of course create L independent linear models, but would like to explore a single model where these sub-linear relationships are partially correlated. The reason being that some levels of c have relatively low statistics and would benefit from the global x~y relationship. Also, we would like a single model when extending to secondary features.

Any suggestions are welcome. (In case of specific libraries, R is preferred.)

$\endgroup$
1
$\begingroup$

A generalized linear model (glm) appeared to be work reasonable.

glm(y ~ x * c, data=df) and extensions hereof.

$\endgroup$

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.