Next: Rule-Selection:
Up: Decision Trees
Previous: Decision Trees
The problem is how to construct a decision tree, how to infer structure
and classification rules from a data set. Most of the construction algorithms
are referred to as Top-Down Induction of Decision Trees (TDIDT)
[45]. Induction, because
the knowledge is acquired inductively from the data-set; top-down because
a candidate rule is chosen first for the single top decision node and then
each of its subsets is recursively partitioned. The splitting is terminated
if all members of a subset belong to the same class or no further decision criteria
are left. Some newer algorithms also stop splitting,
referred to as ``pre pruning'' the tree, if the classification improvement
seems insignificant.
Other algorithms replace insignificant subtrees by leafs in a post process, known
as ``post pruning''.
Thomas Prang
1998-06-07