top of page
  • Writer's pictureTaryn Morris

Forecasting the future with Facebook’s "Prophet".

Forecasting and understanding time based patterns are critical for decision making in many organisations. Simply, time series forecasting allows for the prediction of future outcomes based on past patterns and events.




While the idea behind forecasting is quite simple, in reality, producing a high quality forecast is not easy and requires experience and very specific skills. The entry of the forecasting tool “Prophet”, created, used and shared by Facebook, has however, made the task of creating quality forecasts much easier and quicker and with greater flexibility.


The immediate question that comes to mind might be what does Facebook get from you in return? The answer is nothing. The tool is open source and the full code is freely available for practitioners to use in both Python and R. A useful get started page is available here.


Here are some of the advantages of using Prophet:

  • It can give you an understanding of yearly, weekly, and daily patterns. It can also accommodate custom time periods (e.g. hourly, half hourly, quarterly etc.).

  • It is able to take holidays or custom specified events into account. This is particularly useful for events that do not happen on the same day each year e.g. Easter, Diwali or an event unique to an organisation.

  • It is robust to missing data

  • It handles outliers well

  • It can incorporate large shifts in trends e.g. changes due to system upgrades, new products, unforeseen calamities etc.

  • The parameters are adjustable so the model can be tweaked to get the best forecast possible.

Prophet forecasts generate future patterns based on previous events. In the below example we forecast three months into the future. We can see how the predicted values (blue line) generally match the historical values (black dots) very well. The tool also has built in options for validation of the forecast.

The package also allows for plotting of the patterns over time (often referred to as time series components). In this example we can see that Saturdays have the highest volume of the measured item. We can also see that volumes tend to be highest towards the beginning and end of the month and that December and January see the highest volumes. Overall the volume of the item being measured has decreased from 2016 to 2019. We can also see the effects of holidays on observed patterns.

Prophet is a dynamic, customisable and accurate way to forecast your trends into the future.

Contact us if you would like to unlock the benefits of forecasting for your organisation.


90 views0 comments

Recent Posts

See All
bottom of page