Help System (web edition)

THE CLASSIFICATION ANALYSIS

(1) The classification method used is selected by the METHOD
specification:

METHOD=CLASSIC  (default)
METHOD=BAYES
METHOD=MAHAL   classification based on the Mahalanobis distance only

(2) The classification may be performed by using the classification
functions based on the discriminant analysis (DSPACE=1, default) or
on the original data (DSPACE=2).

(3) The group covariance matrices may be assumed to be equal
(default) or the classification may be done without that assumption
(METHOD=UNEQC).
By combining these three features several formulas for forming
the classification scores can be obtained. By default, the prior
probability that a case belongs to a group is assumed to be
proportional to the sample size. The user may give his own prior
probabilities by the PRIOR specification, e.g. PRIORS=0.25,0.5,0.25.

The program classifies each case into the group with the highest
posterior probability. By default, the results are presented in a
summary table. Casewise classification results may be obtained by
the LIST specification. For each case the printout contains the
Mahalanobis distances and posterior probabilities for belonging to
each group:

LIST=ALL      All observations
LIST=INCORR   Only missclassified observations are reported
LIST=i,j      The printout starts from i'th observation and ends
              with the j'th observation.
The scores of the discriminant functions for each case may be
saved in the Survo data file by giving the names of these new
variables in the CANONICAL specification or they can be pointed
by masks C. The number of these canonical variables is
min(g-1,p), where g is the number of groups and p is the number
of variables used for forming the functions. Only the named
canonical variables are saved. The predicted group may be saved
in the Survo data file by the PREDICTED specification or by mask
P.

If the same data is used for computing the classification functions
and for classifying the cases, then the classification results
may be too optimistic. This may be avoided either by using another
data for classification or by using cross validation methods.
The use of another data file is pointed by the CLFDATA specification,
e.g.

    DISCR FISHER1,END+2
    VARIABLES=sepallen,sepalw,petallen,petalw
    GROUPING=iristype  iristype=1(setosa),2(versicol),3(virginic)
    PREDICTED=prediris
    CLFDATA=fisher2
    CANONICAL=Cano1,Cano2

Note! The new canonical variables Cano1 and Cano2 are saved in
both Survo data files. The predicted group in the data file
fisher2 only.

The cross validation method is used if option CROSSV is stated
in the METHOD parameter and it may be used only if DSPACE=2.
In cross validation, when a case is to be classified the the
the effect of this case is removed from the classification formulas.
Further information:
  1 = Definitions for grouping variables 
  A = More on the discriminant analysis 
  D = More on data analysis