predictive classification/ continuous and binned target variables both included in the training data?


I am using Knime to predict (predictive classification) the annual sales volume 2013 (target variable with three classes) of teams based on background variables and the first three months sales volumes.

The training data includes background variables and sales volume time-series of 12 months in 2012. 

Is it an unsound approach to train the learners with target variable included as binned AND continuous. Obviously, this gives a high accuracy for the learner 2012, since target class can be "predicted" by using the continous target values.

The logic here is that I assume 2012 total sales volume will be a good estimate of 2013 sales volume.

I know this is not a technical question, but I would appreciate if you can share thoughts on this. 

Thank you!


Hi Markus, 

Can you provide an example?  I am having a bit of trouble following your setup.