Outlier Correction
"Outliers" are anomalous values that cannot be explained by the forecast model. They can heavily influence the forecast result.
The system can identify and replace outliers in the historical data. To do this, the forecast procedure calculates forecast values in the past period and compares them to the historical values. If the difference (the residual) exceeds a specific threshold value, the historical value is replaced by the ex-post-forecast value for the corresponding point in time. After this correction has been made, the forecast calculation is performed again using the amended historical data.
To set the threshold value, you can define a Sigma factor.
Integration
Outlier detection depends on the forecast model because the forecast value is calculated using a model and its related algorithm.
Features
In the forecast function, outliers in the historical values are defined by the Sigma factor. The greater the Sigma factor, the more tolerance the system is with regard to anomalous values, meaning that the system will identify fewer outliers.
Sigma factor fac
defines outlier detection as follows: An observed value y
is declared as an outlier if the difference to forecast value e
is greater than fac *s
. s
denotes the standard deviation of the residuals.
Activities
Activate outlier correction if you think that outliers are having an unfavorable effect on the forecast result.
Avoid using outlier correction if you always want to take account of anomalous values when calculating forecasts.