Detecting exceptional values early and accurately is an important step in the forecasting cycle, as faulty data that goes unnoticed can severely disrupt subsequent procedures. Outlier handling is employed to ensure any exceptional values are detected early, so that they can be treated appropriately.

Exceptional values detected by the AdaptiCast outlier handling methods are tagged so that they can be easily identified in further steps. Depending on the application, outliers can be reported to forecasting operators, or they can be treated as missing values to be handled further by automatic methods for dealing with missing values, known as Imputation.

The criteria for outlier detection can be adjusted for the particular purpose. In many cases, domain knowledge is used in combination with statistical properties to identify values that are unlikely to be correct.