Statistics-MVA-BayesianDiscrimination version 0.0.2
Discriminant analysis is a procedure for classifying a set of observations each with k variables into predefined classes such as to allow the
determination of the class of new observations based upon the values of the k variables for these new observations. Group membership based on linear combinations of the variables. From the set of observations where group membership is know the procedure constructs a set of linear functions, termed
discriminant functions, such that:
L = B[0] * x1 + B[1] * x1 +... ... + B[n] * x_n - c
Where B[0] is a constant, B[n's] are discriminant coefficients and x's are the input variables. These discriminant functions
(there is one for each group - consequently as this module only analyses data for two groups atm it generates two such
discriminant functions.
Before proceding with the analysis you should: (1) Perform BartlettĀ“s test to see if the covariance matrices of the data
are homogenous for the populations used (see L. If they are not homogenous you should use Quadratic Discrimination analysis.
(2) test for equality of the group means using Hotelling's T^2 (see L or MANOVA. If
the groups do not differ significantly it is extremely unlikely that discrimination analysis with generate any usefull
discrimination rules. (3) Specify the prior probabilities. This module allows you to do this in several ways - see
L.
This class automatically generates the discrimination coefficients at part of object construction. You can then either
use the C