A comparative study of prediction techniques

This study deals with the techniques for predicting the future demands of spare parts. The results of studies about the Air Force Inventory problem have led us to believe that any improvement in the demand forecast could result in important savings. Demand prediction is actually based upon two separate estimating procedures. First, an estimate is made about the underlying pattern of demands per program element, Second, an estimate is made about the pattern of future program elements. It is the marriage of these two estimates that allows us to make predictions of the number of demands in future time periods. This study is restricted to finding ways of improving the estimates of the demands per program element, We used past demand data with the pattern of program elements known. We also used the known pattern of future program elements to make the prediction of future demand. There are seven prediction techniques used in this study. The first four techniques are based on the assumption that demand rate (demand per program element) is constant through time. The other three techniques assume that failures (demands) are related to the age of the part. These three techniques are routines for estimating the service Life characteristics of parts. We used as inputs to the study two batches of data: 33 months of demand, data on a sample of 300 B‐52 parts and 26 months of failures of 27 missile components. We made monthly estimates of demand for a year in the future and measured the prediction errors. In the evaluation of the estimating procedures, we used three measures of the accuracy of the forecasts, namely average monthly error, relative error expressed as percent of the actual demand and root mean square of the errors of the predictions.