forecast Package
Time series forecasting methods.
Basic Forecasting
library(forecast)
# Create time series
ts_data <- ts(data, frequency = 12, start = c(2020, 1))
# Auto ARIMA
fit <- auto.arima(ts_data)
fc <- forecast(fit, h = 12)
plot(fc)
# ETS (Exponential Smoothing)
fit <- ets(ts_data)
fc <- forecast(fit, h = 12)
ARIMA
# Automatic
fit <- auto.arima(ts_data,
seasonal = TRUE,
stepwise = FALSE,
approximation = FALSE
)
# Manual
fit <- Arima(ts_data, order = c(1, 1, 1), seasonal = c(1, 1, 1))
# Diagnostics
checkresiduals(fit)
ETS Models
# Automatic
fit <- ets(ts_data)
# Specific model (A=Additive, M=Multiplicative, N=None)
fit <- ets(ts_data, model = "MAM") # Multiplicative error, Additive trend, Multiplicative season
Other Methods
# Theta
fit <- thetaf(ts_data, h = 12)
# TBATS (complex seasonality)
fit <- tbats(ts_data)
fc <- forecast(fit, h = 12)
# Neural network
fit <- nnetar(ts_data)
fc <- forecast(fit, h = 12)
# STL decomposition + ETS
fit <- stlf(ts_data, h = 12)
Accuracy
# Train/test split
train <- window(ts_data, end = c(2022, 12))
test <- window(ts_data, start = c(2023, 1))
fit <- auto.arima(train)
fc <- forecast(fit, h = length(test))
accuracy(fc, test)
Cross-Validation
# Time series CV
errors <- tsCV(ts_data, forecastfunction = function(x, h) {
forecast(auto.arima(x), h = h)
}, h = 12)
sqrt(mean(errors^2, na.rm = TRUE)) # RMSE