forunco: Forunco combination approach

In yvesmauron/univariate-time-series-forecasting: Intelligent Algorithms to be Integrated in Business Intelligence solutions.

Description

Combination of univariate time seris methods, using a combination operator of choice for the mean forecasts and prediction intervals.

Usage

forunco(
  y,
  h,
  levels = c(95),
  methods = c("auto_ets", "auto_arima", "auto_thetaf"),
  pp_methods = c("boxcox"),
  point_combination = "median",
  pi_combination_upper = "median",
  pi_combination_lower = "median",
  pool_limit = length(methods),
  error_fun = "rmse",
  weight_fun = "inverse",
  val_h = h,
  sov_only = F,
  max_years = 30,
  val_min_years = 4,
  cv_min_years = 5,
  cv_max_samples = 3,
  allow_negatives = F,
  remove_outliers = F,
  ...
)

Arguments

y	Time series object with historical values.
h	Horizons to be predicted.
levels	Prediction interval levels. Optional, default c(95).
methods	Methods to be combined. Optional, default auto_ets, auto_arima, auto_thetaf.
point_combination	Point forecast combination operator. Optional, default meidan.
pi_combination_upper	combination operator of the upper bound of the prediction intervals. Optional, default max.
pi_combination_lower	combination operator for the lower bounds of the prediction intervals. Optional, default min.
pool_limit	number of methods selected from the pool to reach the final forecasts. Optional, default length(methods).
error_fun	error function to determine the validation performance Optional, default rmse.
weight_fun	weight function for calculating the definitive weights for combination. Optional, default inverse
val_h	The horizon for the validation samples. Optional, default h.
sov_only	Flag indicating whether only single origin validation should be considered. Optional, default TRUE.
max_years	Maxmium of years to consider during model fitting. Optional, default 30.
val_min_years	Minimum years required to conduct single origin validation. Optional, default 4.
cv_min_years	Minimum years required to conduct cross-origin validation. Optional, default 5.
cv_max_samples	Maximum samples that should be considered during cross-validation, i.e. 3 indicates the algorithm validates from 3 origins. Optional, default 3.
allow_negatives	Flag indicating whether to allow negative values or not. Optional, default FALSE.
remove_outliers	Flag indicating whether to automatically remove outliers form the time series. default FALSE
...	passed to the forecasting functions

Value

combined forecasts of time series y including confidence intervals

Examples

# not run
library(Mcomp)
ts <- M3[[2104]]$x
fcs <- forunco(y = ts, h = 18)
# end not run

yvesmauron/univariate-time-series-forecasting documentation built on March 2, 2020, 12:20 a.m.