What meaning has the weighted sum of a group of variables so that each weight is assigned to maximize the minimum resulting correlation of all these variables to the sum obtained?
1 Answer
$\begingroup$
$\endgroup$
1
This technique is performed when the problem asks to create a variable composed of many highly-correlated independent variables.
Sometimes, when a variable can be linearly predicted from others, the other weights in the model can change in a wrong way, so algorithms like Ridge regression help avoiding this effect.
-
$\begingroup$ Yes, but did you hear anywhere about this technique? and what would be the reason to do this procedure instead of, for example, adding the normalized variables together, or doing PCA and obtaining the predicted value that has a positive correlation with all the variables, yet unevenly distributed among them? $\endgroup$– gabrielMar 17 at 22:23