1
$\begingroup$

I have a dataset that I used to train my ML model for prediction. I created my prediction models using XGBoost and Neural Networks.

Now I want to create another model that can give me the causal relationship between the values that I have. What I actually want to know is which parameters influenced in what manner in predicting my outcome?

I know how the outcome is correlated to some of the parameters by using the Correlation Matrix. But it does not give me causality.

I am working with Python, tensorflow, and Scikit learn

$\endgroup$

1 Answer 1

1
$\begingroup$

It is often not possible to obtain causality from your observed data. Your observed data gives you probabilistic dependencies, but since the same dependency can be created by causal dependency in both directions, as well as through hidden confounding, you cannot deduce cause and effect. That is why people often use interventions to obtain causal information.

What you can do is compute the set of all possible causal graphs that could have created your data. This is done by algorithms like e.g. pcalg or fci. But they only work under certain conditions and are usually implemented in R. There is, however, the Causal Discovery Toolbox, which is a python library.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.