The most logical next step after understanding a time series' features and trends is trying to forecast its future development.
As one would imagine, R provides optimal tools to perform this task.
In this recipe, we will leverage the extremely popular forecast
package by Professor Rob J Hyndman. The package provides an always increasing number of tools for performing univariate time series forecasting.
You can find out more on the package on Prof. Hyndman's personal site at http://robjhyndman.com/software/forecast/.
As stated earlier, the only package needed to perform this recipe is the forecast
package. We therefore need to install it and load it:
install.packages("forecast") library(forecast)
stl()
function to the nottem
dataset:nottem_decomposition <- stl(nottem, s.window = "periodic")
forecast <- forecast(nottem_decomposition,h = 5)
plot(forecast(nottem_decomposition))
Let's take a look at the following graph:
In this plot, you will see a section highlighted in blue. I am sure you have already guessed it; the blue section is exactly what we have been looking for in this recipe, that is, forecasted values.