What you did is building a prediction model that gets evaluated on training set only. It is not a reliable model since it is the output of cheating process, and this explains your optimistic result. Consequently, there is no meaningful contribution here :-)