Maximal Posterior Probability Error Localization

摘要

Given a data base and a system of intervariable constraints that acceptable data records are required to satisfy, it is of interest to determine the specific combination of incorrect components in each failing record. The determination of the components is error localization and it is shown that in theory the statistically optimal means of error localization is maximal posterior probability error location (MPPEL). Properties of MPPEL are stated, and connections to minimal fields to impute (MFI), minimal weighted fields to impute (MWFI), and maximal prior probability error localization are made. Simulation results are presented. The tentative conclusion is suggested that for practical application, MPPEL is not significantly superior to MFI. (ERA citation 07:060680)