This is prerelease documentation for a service in preview release. It is subject to change.

The ARIMA Recipe

The Amazon Forecast Autoregressive Integrated Moving Average (ARIMA) recipe is a popular statistical recipe for time-series forecasting. It captures various standard temporal structures (patterned organizations of time) in the input dataset. It works similar to ARIMA modeling on the Comprehensive R Archive Network website.

How ARIMA Works

Use the ARIMA recipe with a dataset that can be mapped to a stationary time series. The statistical properties of a stationary time series are independent of time. They include the autocorrelations that are constant over time. This type of dataset usually contains a combination of signal and noise. The signal might exhibit a pattern of sinusoidal oscillation or have a seasonal component. ARIMA functions like a filter to separate the signal from the noise, and then extrapolates the signal in the future to make forecasts.

ARIMA Hyperparameters

For information about the hyperparameters used for ARIMA, see the Arima function of the forecast package on the Comprehensive R Archive Network.

Tune ARIMA Models

For information about how to tune an ARIMA model, see the Arima function of the forecast package on the Comprehensive R Archive Network.