03-08-2016 04:19 AM
I don't think all of your message was posted correctly, but as I teach data mining at my university I think I can guess at your question: Do you want to know, when datamining (using regression, decision trees neural net or ensemble) which independent sample (training or validation) should be used to fairly or honestly assess cumulative lift or gain?
If that is your question, validation sample is the answer. Because the predictive model, regaardless of type, is always fitted to the training set, and hte validation data set is truely independent, the lift anfd gain performace of the training set based predictive model on hte validation set is the most fair and honest evaluation of the model's ability to generalise to other data - like the new data yet to come in hte future that you will eventually want to score with your shiny new predicve model.
Did I guess right?