How can I know if a dataseet can be forecasted by Prophet
Hello Community, I have been searching and trying different types of datasheets. Some of them gave me a reasonable ratio of error, but others are really high. For instance this file gave me 112% of error, so the model is useless for this example. ProductName(1).csv
@maxlelyonais I know it has been a while since you posted this question, but I wanted to share a few thoughts regarding Prophet and predictions:
-
You can transform numbers using either the natural logarithm or logarithm base 10. This method has consistently improved Prophet's performance with the data I've worked with.
-
Feel free to experiment with the hyperparameters. For a better understanding of which parameters to adjust and when to tune them, I recommend referring to the documentation found here: https://facebook.github.io/prophet/docs/diagnostics.html#:~:text=Parameters%20that%20can,could%20be%20tuned.
I ran Prophet on your data and achieved a Mean Absolute Percentage Error (MAPE) of 112% using the base model with the raw values you provided. However, after converting the raw target values to natural logs, I improved the MAPE to 93%.