auto_arma_model: ARMA model with tuned order

In sfeuerriegel/caret.ts: Time series models for "caret" package

Description

Creates an ARMA model that is then fitted to the data as a univariate time series. If further variables are specified in the model, it also includess exogenous variables. The order (p, q) is tuned by choosing the one with best fit.

Usage

auto_arma_model(p = 5, q = 5, intercept = TRUE, ...)

Arguments

p	Maximum order of auto-regressive (AR) terms that is tested to find the best fit (default: 5).
q	Maximum order of moving-average (MA) term that is tested to find the best fit (default: 5).
intercept	Boolean value whether to include an intercept term (default: TRUE).
...	Further arguments used when fitting ARMA model.

Details

Variable importance metrics return the absolute value of the coefficients for the exogenous variables (if any).

Value

Model definition that can then be insered into train.

Note

If one desires an ARMA model of fixed, pre-defined order, then one needs to switch to auto_arma_model.

Examples

library(caret)

# without exogenous variables

library(forecast)
data(WWWusage) # from package "forecast"
df <- data.frame(y = as.numeric(WWWusage))

lm <- train(y ~ 1, data = df, method = "lm", trControl = trainDirectFit())
summary(lm)
RMSE(predict(lm, df), df)

arma <- train(y ~ 1, data = df, method = auto_arma_model(), trControl = trainDirectFit())
summary(arma)
RMSE(predict(arma, df), df)

# with exogenous variables

library(vars)
data(Canada)

arma <- train(x = Canada[, -2], y = Canada[, 2], 
              method = auto_arma_model(), trControl = trainDirectFit())

summary(arma)
arimaorder(arma$finalModel) # order of best model

predict(arma, Canada[, -2]) # in-sample predictions
RMSE(predict(arma, Canada[, -2]), Canada[, 2]) # in-sample RMSE

absCoef <- varImp(arma, scale = FALSE) # variable importance (= absolute value of coefficient)
absCoef

plot(absCoef)

sfeuerriegel/caret.ts documentation built on May 29, 2019, 8:01 p.m.