Statistical Pattern Recognition

Abstract

Statistical pattern recognition is concerned with the problem of designing machines that can classify complex patterns. Although statistical pattern recognition is typically viewed as a branch of Artificial Intelligence, such problems frequently arise in the social and behavioral sciences in the course of detecting complex structural relationships in large data sets. Statistical Pattern recognition problems also arise in the course of modeling complex social, behavioral, and neural systems. Most statistical pattern recognition systems consist of three major components. The first component is a feature selection and extraction stage where critical informational features about the data are identified for classification purposes. The second component is a probabilistic knowledge representation for representing the expected likelihood of particular features and the associated losses for making situation-specific decisions. The third component is a decision rule for making classification decisions which minimize an appropriate expected loss function. In many cases, a fourth component may also be required to estimate the probabilistic knowledge representation from ‘training data.’