End-to-end user journeys for time series forecasting models

This document describes the user journeys for BigQuery ML time series forecasting models, including the statements and functions that you can use to work with time series forecasting models. BigQuery ML offers the following types of time series forecasting models:

Model creation user journeys

The following table describes the statements and functions you can use to create time series forecasting models:

Model type	Model creation	Feature preprocessing	Hyperparameter tuning	Model weights	Tutorials
`ARIMA_PLUS`	`CREATE MODEL`	Automatic preprocessing	auto.ARIMA¹ automatic tuning	`ML.ARIMA_COEFFICIENTS`	Forecast a single time series Forecast multiple time series Forecast millions of time series Use custom holidays Limit forecasted values Perform hierarchical time series forecasting Perform anomaly detection with a multivariate time-series forecasting model
`ARIMA_PLUS_XREG`	`CREATE MODEL`	Automatic preprocessing	auto.ARIMA¹ automatic tuning	`ML.ARIMA_COEFFICIENTS`	Forecast a single time series Forecast multiple time series
TimesFM	N/A	N/A	N/A	N/A	Forecast multiple time series

¹The auto.ARIMA algorithm performs hyperparameter tuning for the trend module. Hyperparameter tuning isn't supported for the entire modeling pipeline. See the modeling pipeline for more details.

Model use user journeys

The following table describes the statements and functions you can use to evaluate, explain, and get forecasts from time series forecasting models:

Model type	Evaluation	Inference	AI explanation
`ARIMA_PLUS`	`ML.EVALUATE`¹ `ML.ARIMA_EVALUATE` `ML.HOLIDAY_INFO`	`ML.FORECAST` `ML.DETECT_ANOMALIES`	`ML.EXPLAIN_FORECAST`²
`ARIMA_PLUS_XREG`	`ML.EVALUATE`¹ `ML.ARIMA_EVALUATE` `ML.HOLIDAY_INFO`	`ML.FORECAST` `ML.DETECT_ANOMALIES`	`ML.EXPLAIN_FORECAST`²
TimesFM	N/A	`AI.FORECAST`	N/A

¹You can input evaluation data to the ML.EVALUATE function to compute forecasting metrics such as mean absolute percentage error (MAPE). If you don't have evaluation data, you can use the ML.ARIMA_EVALUATE function to output information about the model like drift and variance.

²The ML.EXPLAIN_FORECAST function encompasses the ML.FORECAST function because its output is a superset of the results of ML.FORECAST.