Time series problems

Similarities

• Both involve a numerical target column.

• Both are commonly used in financial use cases involving sales and monetary forecasting.

Differences

• Time series problems support grouped targets, while regression problems do not (see Components of a time series problem). Grouped scenarios can still be addressed for regression problems by training multiple different models, at the cost of global learning across groups.

• Time series problems support scenarios where you know certain feature variables ahead of time—for example, weather-related forecasts, planned promotional discounts, and whether dates fall on week days, weekends, and holidays. These feature variables are known as Future features.

• For time series problems, data must be indexed by date or datetime on a fixed time interval. Also, different data content is expected and generated during training and predictions (see Preparing a training dataset and Preparing an apply dataset).

• In time series problems, predicted values explicitly correspond to specific dates and times. In regression problems, predicted values may or may not correspond to specific dates and times, but if they do, this association is implied rather than explicitly denoted in output.

• Different algorithms are used (see Understanding model algorithms).