Empirical modeling methods using partial data

Abstract

Methods were developed to calculate empirical models for device error behavior from data sets with missing data. These models can be used to develop reduced point testing procedures for the devices. The partial data methods reduce the prediction uncertainty for test points that have more modeling data available relative to the prediction uncertainty of partial data test points. Simulations show that the prediction uncertainty for full data test points are comparable to the case where the "missing" data are "known." When these methods are applied to real data where the underlying model has changed the improvements are less than the simulations predict.