Tree building phases

A decision tree classifier is built in two phases.

• Growth
  
  • Tree is built by recursive partitioning
  • Form of split used to partition the data depends on type of attribute used in split
  • Split for continuous attribute A are of form : value(A) < x,
    x is a value from domain of A
  • Split for categorial attribute are of form : value(A) { ] X where
    X <= domain(A)
  • Only binary split is considered
• Prune
  Once tree is grown up fully, it is pruned to generalize the tree by removing dependence on statistical noise or variations.