FILTERING DEMAND HISTORY FROM HORIZONTAL, TREND AND TREND SEASONAL DEMAND PATTERNS

The thesis presents a filtering method to be used for a time series data. The filtering process is used to identify any demand entry that is widely different from the other demands which is here called an outlier. The process also finds a replaced value which is likely to be more representative compared to the other demands in the pattern to be used, if necessary, for an outlier. The filtering process improves the reliability of the data before being used in the forecasting process.;The number of demand entries used is in the range of two to five years (monthly data). The filtering process output is the report showing the decision of the process regarding the status of each of the demand. For the demand that is assigned as an outlier, the filtering process also reports the suggested replaced value to be used. In a large inventory system where the forecasts are needed without interruption in order to move forward, this filtering process is applicable to help improve the reliability of the data. In this situation, the outlier is identified and replaced in calculating the forecast but not in the permanent record of item.;The parameter required for this filtering process is the value of the "decision ratio". This parameter is used to control the tightness of the filtering process. This thesis shows how the filtering process works for the horizontal, trend and trend seasonal patterns. The various conditions of demand and outlier characteristics in the demand pattern are simulated to test the filtering process. The filtering performance is evaluated from the test results. The value of the decision ratio is also varied to show the filtering performance at different level of tightness of the process.