Filtering Outliers

Abstract

A primary goal of forecasting is to measure the flow of demands from the history months and project to the future months with a minimum forecast error. A way to enhance this goal is by filtering the history demands to seek out any outlier demands and adjust accordingly. As demonstrated in the prior chapter, outlier demands cause much damage to the forecasts and increase the forecast error. Filtering of the demand history is not an easy process, but is important to yield forecasts with minimal forecast error. Reducing the forecast error will reduce the amount of safety stock needed to run the inventory operation. This chapter shows a way to seek out and adjust any outlier demands from the history months when the demand patterns are of the horizontal, trend or seasonal type. The filtering process takes place just prior to generating the forecasts.

Summary

Prior to forecasting, the history of monthly demands are filtered seeking out any outlier demands , and if found, the outliers are adjusted accordingly. Filtering is not easy, but is important to generate forecasts with minimal forecast error. The filtering methods described here are developed when the demands follow a horizontal demand pattern, a trend demand pattern and a seasonal demand pattern. Another phase of filtering the demands, occurs in the order entry of the inventory location. This is when the orders come in from customers and the purchase order states a part number and quantity. The quantity is verified to be statistically consistent with those from the part’s history of demands. Any line demand detected as an outlier is sent back to the customer for verification.