Technical Articles
Data Cleansing:Outlier Correction Methods in IBP-Demand
Outlier Detection Calculation with IQR Method:
Output generated with outlier value for Actuals Qty Adj KF
Scenario 2: Variance Method Detection and Correction with Tolerance
Input KF – Actual Quantity
Output KF – Actual Quantity Adjusted
Historical period: 11 weeks
Run Statistical Forecast with 11 weeks Historical Method
Conclusion
In this blog, I have explained only the Outlier correction method ‘correction with Tolerance’ and covered new Outlier correction methods for Trend and Seasonality Introduced in the IBP-2011 Version. Other methods of Outlier Correction are all self-explanatory and hence I am not covering them as a part of this blog. So with this, I would like to conclude that any forecasting model if it needs to obtain a better accuracy it requires a defined pre-processing step which should cleanse the complete dataset.
Please post any of your queries in the comment section.
Good summary of the Outlier correction methods in IBP Demand.........
However, these methods are based on a simple mean and use range-based measures around the central tendency to detect outliers.
Since the detection is not based on thresholds (or Tolerance Lane as SAP has often called it) that are calculated with reference to the conditional mean, typically this will overcleanse the data.
Although these methods sound fancy, caution is advised.......