Learning by Building Identification Trees

Learning by Building Identification Trees

1. Identification Tree

   • a special kind of decision tree to represent property-testing procedures
   • one way to recognize situation by applying the sequence of tests dictated by identification tree
   • Occam's razor
     • one way to learn by building an identification tree, keeping it simple in harmony with Occam's razor
     • the smallest identification tree that is consistent with the samples is the one that is most likely to identify unknown object
   • Sprouter
     • procedure to generate an identification tree until each leaf node is populated by as homogeneous a sample set

2. Disorder formula

   • One way to build a simple identification tree by using a disorder formula, borrowed from information theory, to determine which tests to include in the tree
   • Average disorder

3. From Trees to rules

   • Contingency Table
     • Table that shows the degree to which a result is contingent on a property
   • Default rule
     • Rule that is to be used only if no other rule applies
   • Once an identification tree is built, you usually should convert it into a simple set of rules so as to make the knowledge embedded in it more comprehensible. To do the conversion, you make a rule for each path through the tree, and then you simplify the resulting set of rules. To simplify a set of rules, you first eliminate unnecessary rule antecedents using Fisher's exact test. Then, you eliminate unnecessary rules.

Images and files in the map