0
$\begingroup$

I have a database of characteristics of a signal. The characteristics are a bunch of indicators (kurtosis,skewness,etc). The dependent variable is fault. The fault variable has as a value 1 if the signal corresponds to a faulty machine and 0 if it's not a faulty machine.

I would like to know if there is a method or an algorithm capable of generating indicators (new independent variable) from the ones in the database(e.g :the multiplication of 2 of them or division ) that would correlate better with the dependent variable and would consequently a better indicator of the existence of the fault.

Thank you !

NB : If technical terms are not well used or more details are needed, don't hesitate to ask me as I am new to the field.

$\endgroup$
0
$\begingroup$

You want to look at variable interactions. When building your model, consider something more of the form $y = (x_1 + x_2 + \dots +x_n)^2$, which will give you the pairwise interaction terms (higher exponents will reveal interactions between more variables). You can then see if these have low p-values, etc.

Also, since you say you're not up on the technical lingo, what you're trying to build is a binary classification model. You'll want to look into a linear regression linked to a logistic function.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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