Exponential smoothing

The other alternative to ARIMA is the exponential smoothing technique, which is also a time series forecasting method for univariate data, where random noise is neglected, revealing the underlying time structure. Although it is like ARIMA in that demand forecast is a weighted sum of past observations, the method of applying weights to lagged observations is different—instead of providing equal weights to past observations, the model employs exponentially decreasing weights for lags. In other words, the most recent observations are more relevant than historical ones. Exponential smoothing is used to make short-term forecasts, where we assume that future patterns and trends will look like current patterns and trends.

Get Hands-On Artificial Intelligence on Amazon Web Services now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.