If I normalize training data then apply the gradient descent algorithm which outputs hypothesis $h(x_1,...,x_n)$, I understand that I need to also apply the same corresponding transformations to the $x_1,...,x_n$ of any new test data before plugging into h if we'd like to get predictions for this test data. My question is - what is the interpretation/use of this predicted value, given that the inputs have been transformed and the output prediction we are interested would likely be one computed from non-transformed inputs? (E.g. If the output we are predicting using the model is house pricing, a prediction of $y=7$ is of no practical use; it seems to me that an actual prediction of housing price, like $y=750k$, would be more valuable.)
Thanks