Hybridization of Decision Trees with other Techniques - Data Mining with Decision Trees: Theory and Applications - page 250

concerned it can be pruned to a leaf or its test can be altered. Specifically,
there are four modification options in case of non-leaf node:

• A completely new test is applied with another randomly chosen attribute

and threshold.

• A new threshold is randomly chosen without changing the attribute used

in the test.

• The current sub-tree is replaced by a sub-tree of an adjacent node.

•

The test can be exchanged with another test taken from randomly chosen
descendant-nodes.
If a leaf node is to be mutated, then there are two options:

• The leaf node is replaced with a non-leaf node with a new randomly

chosen test.

• The leaf node is replaced with a sub-tree generated using an appropriate

algorithm.

The CrossTrees operator is equivalent to the standard crossover
operator. It alters two solutions by exchanging certain parts of input trees.
There are three possible exchange types: two types of sub-tree exchanges
and an exchange of only tests. At the beginning, regardless of the type, one
node in each tree is randomly chosen. Then the type of exchange between
trees is decided.
In the first CrossTree variant, the tests that are associated with the
chosen nodes are substituted. This option is valid only when the chosen
nodes have the same number of outgoing branches. In the second CrossTree
variant, we substitute the sub-trees starting from the chosen nodes. The
third CrossTree variant actually combines the first two variants. Branches
which start from the chosen nodes are exchanged in random order.