Facebook Prophet with configurable growth models, multiple seasonality patterns, holiday effects, and robust date parsing.
Configurable yearly, weekly, daily patterns with additive or multiplicative modes.
Linear, logistic, or flat growth with adjustable changepoint flexibility.
Custom holiday definitions with configurable windows before/after each event.
Provide time_column and value_column (automatically converted to ds and y). Supports multiple date formats with intelligent parsing.
Includes component decomposition (trend, seasonalities, holidays), changepoint detection, residual analysis with Q-Q plots, and configurable prior scales for all components.
From components to calibrated predictions
Robust date parsing with 12+ format attempts; convert to Prophet's ds/y format.
Set growth model, seasonality modes, prior scales; add regressors if specified.
Extract components, calculate MAE/RMSE/MAPE, generate residual diagnostics.
Facebook Prophet with enterprise-grade flexibility—handles missing data, outliers, and complex seasonal patterns automatically.
The implementation provides configurable prior scales for trend changepoints, seasonality strength, and holiday effects. Supports additional regressors with automatic future value imputation (uses mean of last 30 days). Includes comprehensive residual analysis, actual vs predicted plots, and simplified seasonal pattern visualizations for yearly, weekly, and daily effects.
Note: Handles 12+ date formats automatically. For regressors in forecast period, uses mean of last 30 historical values. MCMC sampling optional for uncertainty quantification.