Men you are doing completely bullshit. Your script severely overfitting even the way you try to split your test and training set is incorrect. You try to use fancy term in order to persuade your reader that you understand what you have done. Please refresh you statistical knowledge.