When I do a linear regression, R²: 0.90, but the estimates are not correct, why is this happening?
(Deep Not : Adjusted R-squared: -0.3872)
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
You should also check the correlation between the features. The problem you mentioned arises when features are highly correlated.
Also perform Kfold validation in your dataset.
answered Aug 11, 2018 at 18:05
-
$\begingroup$ thanks,Do you have a suggestion for feature selection? $\endgroup$ Aug 12, 2018 at 8:51
-
$\begingroup$ You should really spend some time in selecting your features as it will yield better results. What I tend to do is first plot the distribution plot of each features. If it closely resembles a normal distribution curve then such features may be important for you. Sometimes you may have to use feature scaling and transformations if the distribution is skewed(such as log transformations, boxcox, etc.). After performing scaling or transformation to these variables, there is a high chance that the skewness is removed. Then you can include these transformed variables in your model. $\endgroup$ Aug 13, 2018 at 17:26