DISCRIMINANT ANALYSIS IMPLEMENTATION BASED ON VARIABLE PREDICTIVE MODELS FOR SIMILARITY PATTERN CLASSIFICATION

Abstract

At present, there are many pattern classification methods that can be used such as
LDA, kNN, Bayesian networks, CART, ANN and SVM. However, many classification
methods mentioned above causes some issues. The problem are the large computational
cost, and weakness of methods mentioned above to classify the class because it is based
solely on inter-class boundary (decision boundaries).As an alternative method other
than the methods have already been exist, the relationship between features (interrelation)
in a class can be used to classify a sample of a particular class. Based on
these ideas variable predictive model method based class discrimination (VPMCD) is
proposed by (Raghuraj &Lakshminarayanan, 2008) as a new classification approach to
the problem of large data and overlapping which cannot be easily solved by the other
classification methods.The testings are done using six well studied data sets (Diabetic,
Hear, Iris, Wine, Digit, Letter). The results are equations wihich have capability to
clasify new sample.