I use xgboost in Rstudio to successfully predict daily absolute variables, like sales, clicks, renewals etc. However, if I use exactly the same model to predict for example instead of renewals, with R^2=93%, the renewal rate= renewals/total_count, all my models suddenly don't work anymore, and R^2 <0. I have seen the same trend every time I try to predict a rate, so for example I can predict absolute clicks with 78% accuracy and sales with 83% accuracy but when I try to predict the rate sales/clicks R^2 goes to -300. Clearly, I am doing something wrong when approaching rates, this is more of a conceptual question so,
- is the case of scaling the dependent variable which as a rate stand between 0-1 as a float while other numeric variables go can take all the values as integer 0-3000?
- Is it the case that these rates move very little, in this case between 0.5034710 to 0.5883085?
has anyone faced this before and is able to share suggestions, please?