I currently have a model that has several numeric Y or predicted variables
I am currently using regression (multioutput regression with gradient boosting) to predict the Y variables.
Currently I am not happy with the accuracy and was looking at ways to increase accuracy. I was recommended using discretization so I treat Y as bins or quintiles. i.e:
|2000 - 3000||0.3 - 0.4||...||30000 - 40000|
|1000 - 2000||0.5 - 0.6||...||51000 - 50000|
How can I go about doing this? What algorithms or libraries perform this? I am finding online how to discretize features but not so much the data itself. Do I need to manually shape the trained data into bins and then train on it?
I guess what I'm asking is where to go from here coding wise.