Skip to content
Time Series & Forecasting

Classical Forecasting

Prophet & the Automation Era

8 min+75 XPStep 1/10

Forecasting as decomposable curve-fitting

Prophet (Meta, 2017) recast forecasting as additive regression: y(t) = trend + seasonality + holidays + noise, fit by ordinary optimization rather than classical time-series machinery.

Its ingredients are pragmatic:

  • Trend: piecewise linear (or saturating logistic) with automatically detected changepoints — it bends where the world bent
  • Seasonality: smooth Fourier terms for yearly, weekly, and daily patterns
  • Holidays: a simple calendar of named days with learned effects

It tolerates missing data and outliers, needs only a two-column dataframe (ds, y), and ships with defaults meant for people who have other things to do.

insight

Prophet's real innovation wasn't accuracy — benchmarks put it near, not above, classical methods. It was accessibility: knobs that match analyst intuition instead of ARIMA orders requiring a statistics degree.