Classification

 

The purpose of classification is two-fold. First, we want to test generated rules against another dataset, and see how good the classification is. In addition, we would like to take data sets without the value for the decision attribute and classify each object in the best possible manner.

In this chapter we concern ourselves with using already created rules in the best manner for classifying objects. This situation is shown in Figure . Rules created from a training set by a rule generation program is used by a classification module to classify new objects. How the rules have been made will be taken into account, but we will assume they have been created in the best possible manner, and will not consider other ways of creating them.

  
Figure: Classification process.

In the first section of this chapter we take a look at how rules may be used for classifying objects. Then we discuss the rules available from the system RGEN which creates rules and the opportunities these give, before we discuss what we actually have implement in the last section.

• Methods for Using Existing Rules
  • Accuracy of the Rules:
  • Weight of Evidence Measure
    • Detection of Underlying Patterns
    • Rule Generation
    • Determination of Class Membership
• Rules Available to the Classifier
  • Problems and Opportunities
    • Conflicting Rules
    • Rules in a Sub-Lattice
    • Multiple Equal Rules
• The Implemented System
  • Testing or Classification of New Objects
  • Statistics
  • Validity Measure
  • Methods
    • First Rule
    • The Rule with Highest Accuracy Matching
    • Simple Voting
    • Voting by the Weight of Evidence Measure
  • Ignoring Blocks
  • Using Certain Attributes
  • Using Certain Rule bases
  • Possibility for Later Additions