Please use regression instead of classification and compare results

[harveybc]

Hi, maybe a silly question but, if the data to be predicted is ordered (such as the BCS score(no matter it
is discrete)), maybe regression is a best fit, since using classification if an error occurs you don't have guarantees that the selected class is the nearest to the real class, and the magnitude of the error (distancce to real data) maybe of use for your regression training algo which does not happen in classification. I.E if you missclassify a 3 it can be missclassified as 1 or as 4.

That may mess both your training algorithm and your metrics.

Hope it helps. Sayonara.