SAS学习资料 Discrimination and Classification.pptVIP

Chapter 11 Discrimination and Classification 11.1 INTRODUCTION Discrimination and classification are multivariate techniques concerned with separating distinct se-ts of objects (or observations) and with allocating new objects (observations) to previously defined groups. Classification ordinarily requires more pr-oblem structure than discrimination does. Thus, the immediate goals of discrimination and classification, respectively, are as follows: Goal 1. To describe, either graphically (in th-ree or fewer dimensions) or algebraically, the diff-erential features of objects (observations) from se-veral known collections (populations). We try to find “discriminants” whose numerical values are such that the collections are separated as much as possibles. Goal 2. To sort objects (observations) into two or more labeled classes. The emphasis is on deriving a rule that can be used to optimally assi-gn new objects to the labeled classes. We shall follow convention and use the term discrimination to refer to Goal 1. This terminolo-gy was introduced by R. A. Fisher [9] in the first modern treatment of separative problems. A more descriptive term for this goal, however, is separa-tion. We shall refer to the second goal as classific-ation or allocation. 11.2 SEPARATION AND CLASSIFICATION FOR TWO POPULATIONS To fix ideas, let us list situations in which one may be interested in (1) separating two classes of objects or (2) assigning a new object to one of two classes (or both). It is convenient to label the clas-ses and . The objects are ordinarily separated or classified on the basis of measurements on, for instance, p associated random variables . We can think of the totality of values from the first class as being the population of values for and those from the second class as the population of values for . These two populations can then be described by probability density