tsforecast_tbats | Fits TBATS (trigonometric seasonality, Box-Cox, ARMA errors, trend and seasonal components) to one series (y) or every series of a panel (data), for one or several seasonal periods that may be long (e.g. [7, 365] for daily data). | |
tsforecast_mfles | Fits the gradient-boosted decomposition model to one series (y) or every series of a panel (data): rounds of median, Fourier seasonality (several periods), piecewise-linear trend with changepoints, exogenous regression and smoothing, each shrunk by its learning rate; auto_mfles picks the settings by a backtest search. | |
tsforecast_mstl | Multiple seasonal-trend decomposition by LOESS (MSTL) of one series (y) or every series of a panel (data) for one or several periods (e.g. [7, 365] for daily data, [24, 168] for hourly), forecasting each seasonal component by seasonal naive and the trend plus remainder with a chosen non-seasonal model (automatic ETS by default, or ARIMA, Theta, CES, …). | |
