I am using a model that forecast predictions for DAUs (daily active users). The DAU dataset is seasonal, so I'm trying to figure out the right "error" function for my model.
(The model I'm using Prophet https://facebook.github.io/prophet/ )
The green line is the prediction, the black line is the real data raw.
I am using relative error to calculate the error rate. The relative error formula is: (prediction - real)/prediction.
What's obvious here is that the formula will weight the troughs a lot more than the peaks, since the trough has smaller values.
What are the typical ways to account for this in the formula? I asked chatgpt and I got the following response https://pastebin.com/iEhNfP3q
Some of it makes sense, such as converting the values to logarithmic scale. I'm wondering if there is a "standard" way to handle such cases though.
I'd appreciate some guidance, thank you!