2
$\begingroup$

I would like to generate a regression based on 6 different inputs and 4 output.

I use python and I compare sklearn and scipy. But in both cases, regression models are essentially focused on 1 output parameter.

My question is : can I considerer that all of my outputs are unlinked, and if yes, is it true to imagine that I can perform 4 regressions in parallel (one for each output) to make an equivalent 4 ouput model ?

Tank you for your tips !

$\endgroup$
1
$\begingroup$

"My question is : can I consider that all of my outputs are unlinked"

No, in general you can't, unless you have prior knowledge indicating so. There is a high chance that your outputs are correlated. To explore linear correlations, perform a correlation analysis (calculate correlation matrix). To investigate nonlinear relations, calculate the distance correlation.

and if yes, is it true to imagine that I can perform 4 regressions in parallel (one for each output) to make an equivalent 4 ouput model ?

If your analysis proves that there is low correlation between the outputs, only then you can implement 4 independent regression models in parallel with meaningful results.

My suggestion would be to implement a MIMO machine learning model (multiple inputs / multiple outputs), such as a multivariate Recurrent Neural Network.

Hope it helps :)

| improve this answer | |
$\endgroup$
0
$\begingroup$

I found something on sklearn to generate a multi output from simple output regression. here it is: http://scikit-learn.org/stable/modules/generated/sklearn.multioutput.MultiOutputRegressor.html

It looks to be a superposition, isn't it ?

| improve this answer | |
$\endgroup$

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.