An approach to forecasting that breaks down the underlying pattern of a time series into cyclical, seasonal, trend, and random subpatterns. The subpatterns can then be analyzed individually and recombined to obtain forecasts of the original series.