Prophet

Prophet is a Time Series Forecasting method developed by facebook for modeling time series data with trend, seasonality, and holiday effects. It is designed to handle business time series with strong seasonal patterns and missing data.

How it Works

Prophet models a time series as an additive combination of components:

$y(t)=g(t)+s(t)+h(t)+{ϵ}_t$

where:

• $g(t)$ - trend component (piecewise linear or logistic growth)
• $s(t)$ - seasonal component (modeled with Fourier series)
• $h(t)$ - holiday effects
• ${ϵ}_t$ - error term (irregular noise)

It fits these components using curve fitting and regression, rather than classical statistical assumptions (e.g., stationarity).

Benefits

• Interpretable: Components (trend, seasonality, holidays) are explicitly modeled and easy to visualize.
• Robust: Handles missing data, outliers, and irregular time series.
• Flexible: Captures both additive and multiplicative seasonal effects.
• Automatic: Minimal tuning required - works well with default settings.
• Scalable: Efficient for large datasets and multiple time series.