Dear R community,
I’m new in R (coming from SAS). If possible, I’d like to know your thoughts
regarding the following.
I’m working on a regression problem, where my dependent variable is a ratio
of two variables (i.e. dep.var=A/B). The variable A takes a value of zero
for the vast majority of observations and, of course, so does the dependent
variable. I thought I should do a weighted analysis (weighting on the
denominator of the ratio).
I could accomplish this with a GLM (using a Poisson ~ for the dependent
variable with weights=B) . However, I’d like to try a data mining approach.
Specifically, I’m thinking of using a Neural Network with activation
function for the target layer=exponential and error function=Poisson with
weights=B. Is there any package that would allow me to do this?
Alternatively, is there any other package with an algorithmic approach (as
opposed the classical statistical analysis) that would allow me to
accomplish what I want do?
Thanks in advance for your help!
Axel.
[[alternative HTML version deleted]]
