I’m not so sure why you need to transform the ‘y’ vector ? I assume by ‘Y’ you meant the labels, right ? In a supervised problem ‘Y’ (label) will be your prediction, so why you want to train a model with your potential outcome transformed ?