An analysis of the prediction accuracy of the U.S. Navy repair turn-around time forecast model

Abstract

This thesis examines the forecast accuracy of repair times for a subset of repairable U.S Navy inventory items. Forecasts are currently calculated using the Uniform Inventory Control Program (UICP) on a quarterly basis. The UICP model use the time of repairs completed in the current quarter to update a "file" value in order to forecast the repair times for the following quarter. Forecasts are calculated separately for repairable items grouped into families. This thesis demonstrates that aggregation repairs by their completion dates, as currently done by the UICP model, causes forecast to be affected by the nature of the repair arrival process. The more that this process differs from a Poisson process, the more that the forecast values are affected. Using bootstrap simulations, the effect of the repair process on the forecasting is quantified. This thesis also explores alternatives to the UICP model for forecasting repair times. In particular, an approach that utilizes repairs that have not been completed by the end of the current quarter is developed.