Response variable has different values for same set of predictors I am working on a regression problem for a dataset which has different values of response variable for the same set of predictors. Also, my response variable has a wide range of values starting from the value 50 to 5000 and higher.
For example:
    col1  col2  col3 Response
1    A     B     C     63
2    A     B     C     4354

I have log transformed the Response variable to reduce the effect of the wide range.
I am using RandomForestRegressor in Python for prediction. I have got a $R^2$ value of 0.87 but my model is not giving values close to the original values, I am guessing that is because I have transformed the data.
Can somebody please guide me on how I can model such data?
Please also let me know if I haven't made myself clear.
Thanks.
 A: I don't know a lot about random forests, but generally speaking, if the same predictors have widely varying outcomes, it is one of several scenarios:


*

*There response is random. That is to say, there is no correlation between the predictor variables and the outcome variable. If this is the case, any transformation you do will just be hand waving and will not get you anywhere unless you find the unaccounted variables that do affect the variance in the output.

*There is a relationship between the predictors and the output, but it is mediated by some other variable (that is, it can be that the output can be small or large, based on some unaccounted for variable, but the variance within the large or small categories is explained by your current predictors). Again, you can't wave this away without more data.

*The data is corrupt. If there is a small amount of large or small values (or both), than these outliers might be bad data (or one of the two points above). Try plotting the distribution of your data. see how many outliers you have.


Besides this, 

but my model is not giving values close to the original values, I am guessing that is because I have transformed the data.

Sounds right.
