If I am creating a multiple regression model such as Y = 3 + 1.5apple + 2melon and some of the indipendent variables underwent a transformation as they were not linear or they showed heteroskedasticity, how do I represent the indipendent variables' coefficient in the model?

Let's say that the equation above is made by the intercepts (3) and that the first variable or "apple" was transformed through a log10 because its residuals were strongly positive skewed:

A. Should I report its coefficient in the equation as 1.5log(apple)?

B. If so, how should I report variables transformed by "Sqrt(Variable)", i.e. the transformation that happens when data are moderately positively skewed? Should I just report the variable as "1.5apple^2" ?

C. How do I report variables transformed to deal with negative skewing? When data are strongly negative skewed the operation to transform the variable is log10(Max Value of Apple - Apple). How do I report into a regression a variable transformed in this way? Let's say the maximum value of apple is 3 should I report this variable such as "1.5log(3 - Apple)"? ]]>

Can someone provide more insights on how we can achieve forecasting with fewer data points and come up with more accurate models.

I have tried regression however MAD is at 18% and is too high for my model.

Thanks! ]]>