In an ongoing project - where we are tasked with modelling a complex organization - we were recently made aware of the "Prophet model": an opensource model released by Facebook that is apt for modeling time series data with seasonality.

Prophet provides two enhancements we find interesting.

First, it makes it easy to add "changepoints": discrete events that have a strong impact on your output depending on what you are trying to model. For example you would want to account for public holidays or the Superbowl if you were trying to model traffic.



Second, it makes it straightforward to break down the seasonality components of the forecast into a daily, monthly or yearly view. It is possible to do address these points in ARIMA or KNN, but Prophet does make it easier.



We're going to be experimenting with it a lot more in the coming days and pit it up against our usual favorites. If you are interested in more information we recommend Facebook's posts on the topic, two examples located here and here.