Hello,
I was wondering if you could further explain why MAPE is a better metric to be used in California housing data than MAE?
MAPE is a relative metric while MAE is not. Let’s give a concrete example.
Imagine that you are selling houses with prices ranging from 50 k$ to 150 k$. Having a mean absolute error (MAE) of 50 k$ is really problematic. It corresponds to the price of the house or to 1/3 of the price of the house for the most expensive house.
Now, let’s imagine that you are selling houses with prices ranging from 500 k$ to 1 M$. Having an MAE of 50 k$ is not as dramatic as in the previous case. Intuitively, MAPE makes the MAE relative depending on the actual of the price house that you try to predict. So making a 50 k$ error on a house valued at 50 k$ is an error of 100% while 50 k$ on a house valued at 500 k$ corresponds to a 10% error. The MAPE would average these absolute percent error.
1 Like