Identifying a compact, informative set of input variables is an important factor in achieving high forecasting accuracy, and can have a large positive impact on forecasting performance. The process of selecting a subset of the available input variables is called feature selection. While feature selection can be done by hand, the number of available features is often large, and the resulting number of possible feature combination is astronomical.

AdaptiCast Feature Selection technology provides automatic methods to perform feature selection. By analyzing the informative value of large amounts of input variable combinations, AdaptiCast Feature Selection can identify compact informative sets of input variables that often constitute improvements compared to variable combinations chosen by experts.

The Feature Selection process employs information-theoretic measures of the amount of information different features capture about the target data (the data to be predicted). Furthermore, non-linear contribution of different feature combinations are taken into account. Together, the principles employed by AdaptiCast Feature Selection are able to identify compact yet informative subsets of the available input variables.

AdaptiCast Feature Selection technology has established significant improvements in forecasting accuracy for Adapticon customers, leading to strongly reduced costs. See whether your forecasting performance can be improved by leveraging AdaptiCast Feature Selection.