Outlier adjustment refers to the elimination of unusual peaks or troughs in demand in a historical time series from which a statistical demand forecast is to be calculated.
An outlier adjustment can be made at three points in the forecasting process. Directly when the data is transferred to the forecasting process (preprocessing), through statistical correction of outliers or through manual corrections. In preprocessing, different data sources can be evaluated together in order to eliminate any impurities in the time series in a differentiated manner.
Statistical outlier correction uses various algorithms to cut values in a time series from a certain peak value or to raise values below a minimum value to this level.
Manual outlier correction is very time-consuming and therefore often unrealistic.
Our tip:
Preprocessing offers the most sophisticated automatic mechanisms for outlier elimination. However, this requires that you have specific information on how to correct the data. If this is the case, you should first check the possibility of preprocessing, but this process cannot normally be carried out by the user himself, as additional programming is usually required. If specific information on the adjustment of outliers is missing, anomalies can be recognized and eliminated by the adjustment algorithm using statistical outlier adjustment. In this case, test different settings for the cleanup parameters to find the best ones. This can be done particularly efficiently and automatically using simulation.
Picture: ©Manuel Pfeiffer/aboutpixel
