2
$\begingroup$

I have data that has been grouped into 27 groups by different criteria. The reason for these groupings is to show that each group has different behavior. However, I would like to normalize everything to the same scale. For example, I would like to normalize to a 0-1 scale of 0-100, that way I could say something like $43^{rd}$ percentile and it would have the same meaning across groups. If I were to just, say, standardize each individually by subtracting the mean of each and dividing by standard deviation, would this work? Would I have to calculate the mean/standard deviation of all of the combined data or do each of the 27 groups individually?

$\endgroup$

1 Answer 1

0
$\begingroup$

You can normalize each criteria independently in values between 0 and 1 without taking into account the other criterias, it will work better for most classification methods k-nearest neighbors, random forest, neural network, etc.

$$x^*_{i,j}=\frac{x_{i,j}-x^{min}_j}{x^{max}_j-x^{min}_j}$$

$\endgroup$

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.