Stability of Classification in Sequential Allocation Analysis

摘要

The present paper is devoted to the analysis of some problems arising in sequential pattern recognition when a stopping criterion is considered.The problem of the reliability of the classification and of its “stability” when new measurements are performed is studied introducing the “probability of reversal”.The relationship between this probability and more familiar ones (probability of belonging to one out of m classes, probability of error of first and second kind) is outlined.Numerical data concerning a medical problem (various forms of liver diseases) are presented.