predict_sales: Creates forecasts based on historical data

In Upsylon/predictor: Creates forecast based on historical data

Creates forecasts based on historical data

Description

This function enables to create forecasts based on historical sales data. It considers a multitude of models and selects the model that had the best performance on the testing set.

Usage

predict_sales(sales_data, frequency = 52, start = c(2014, 1),
  end = c(2019, 12), forecast_horizon = 52, size.te.set = 52,
  promo_done = FALSE, future_impactors = NA, criterion = "accuracy",
  sizeroll = 11)

Arguments

sales_data	A vector containing historical sales data.
frequency	A numerical value specifying the frequency of the seasonality.
start	A vector of length 2 with the date of the first observation. It contains first the year and then the day/week/month according to your data.
end	A vector of length 2 with the date of the last observation. It contains first the year and then the day/week/month according to your data.
forecast_horizon	An integer value specifying the number of observations to forecast.
size.te.set	An integer value specifying the size of the testing set.
promo_done	A logical variable specifying if promotions are done for the product.
future_impactors	To be inputted when using a dynamic model (when promo_done == TRUE). A matrix composed of multiple vectors. For promotions, the vector is composed of binary variables that are equal to 1 when there will be a promotion in the forecasting horizon and to 0 otherwise. Other vectors can also be inputted to take into account other impactors.
criterion	A string variable specifying the selection criterion that should be used to select the model ("ME", "RMSE", "MAE", "MPE", "MAPE", "MASE", "ACF1", "Theil's U"). "accuracy" can also be used to reflect the needs of the company.
sizeroll	The window of the moving average or moving median when using the baseline() function.

Value

A list containing the select model, the associated graphs, the predictions and the confidence intervals, the accuracy measures and the same elements for all other considered models.

Author(s)

Grandadam Patrik

Examples

data("mydata")
my_predictions <- predict_sales(mydata, promo_done = TRUE, future_impactor = c(0,1,0, rep(c(rep(0,6),1), 7)))
my_predictions
my_predictions$selected_model$PLOT # the plot of the selected model
my_predictions$selected_model$FORECAST # the forecast of the selected model
my_predictions$all_models$arima # all components of the arima model

Upsylon/predictor documentation built on June 16, 2022, 4:48 p.m.