Sep-28-2021, 09:05 AM
I have two features that have a strong correlation (> 0.7). Because of this, it's not clear, which of them has a stronger influence on the result: during the fitting, the estimator can include the impact of the correlated part in any of the features and give an incorrect result. Most likely, he will not even include the correlated part in one of the features, but will arbitrarily divide its influence between the features.
I would like to represent two original features in the form of three features: a common part, and two uncorrelated residuals. This way I can see the effect of the residuals on the result.
How to perform such a feature- presentation?
I would like to represent two original features in the form of three features: a common part, and two uncorrelated residuals. This way I can see the effect of the residuals on the result.
How to perform such a feature- presentation?