Business Forecasting with Facebook’s “Prophet”

Virtually every business decision and process is based on a forecast. A company uses its past sales data to forecast what its sales volume will be for the next 12 months. The forecasts help the company to allocate what resources (such as employees) might be required to meet the demand.

Business forecasting can be very challenging. First, the time series itself may require specific domain knowledge about the business such as any particular spikes in the past years. Second, there exists a huge list of quantitative forecasting methods that may become additional burdens for data analysts. Third, the forecasting process should be fast and interpretable.

In this article I will demonstrate the “Prophet” module, open-sourced by Facebook.com, to produce forecasts for planning and goal setting. It is available in both R and Python on Github. “Prophet” is easy to use.

I have also written tutorial posts on NeuralProphet, a successor to Facebook Prophet. You are invited to read

• NeuralProphet (I) — Trend + Seasonality + Holidays and
• NeuralProphet (II) — AR + Lagged and Future Regressors.

Great. Let’s start with the fundamentals.

Behind the Scene — the Generalized Additive Model (GAM)