Seasonal Auto-Regressive Integrated Moving Average (SARIMA)


Citekey	GreisEtAl2018Comparing
Source Code	https://www.statsmodels.org/stable/generated/statsmodels.tsa.statespace.sarimax.SARIMAX.html#statsmodels.tsa.statespace.sarimax.SARIMAX
Learning type	unsupervised
Input dimensionality	univariate

Dependencies

The (p,d,q,P,D,Q) orders of the SARIMA model are automatically determined using statistical tests and stepwise refinement (grid search). You can overwrite this tuning behavior by supplying your orders to fixed_orders, e.g. fixed_orders = { "order": (2, 0, 3), "seasonal_order": (0, 0, 2) }. The period m is automatically added.
Using exhaustive_search=True, the orders are searched for using a grid search without any prior statistical tests. This drastically increases runtime, but finds the optimal model.
The point anomaly score is the absolute error between forecast and original value.
We use SARIMA in an iterative way, fitting model on the first train_window_size points, forecasting forecast_window_size points, and re-calibrating the SARIMA-parameters after each prediction.
If max_lag is set, then the order of the SARIMA model is retrained after max_lag points before making further predictions.