Transformation is required to normalize the output variable.

If you do not transform "y" then you see a different regression line which is not optimized and hence we will get incorrect predictions..

I hope this explanation helps!

Thanks!