|
|
Decision Tree Node |
A basic implementation of a decision tree algorithm has been included as a method of performing a complete classification (opposed to the SA which performs partial classification). A wizard is available to guide the user through using the tree in training and evaluation modes. After constructing the model in training mode, the tree is displayed in one of three formats: standard (using a Windows tree control), formatted XML (using an XSL style sheet) or raw XML. If the classes in the database are of varying size, e.g. class YES has 100 records compared to 900 in class NO, it is common to use a balancing node in the stream before the decision tree node
The decision tree node is used to train, test and view decision trees that can be used for complete classification tasks. Unlike the discovery node which targets a particular class of records, a decision tree will try to build rules to classify all records.
The decision tree algorithm recursively partitions the training data until either the majority of cases belong to a single class, or there are too few records to split. At this point a leaf node is created and assigned the dominant class.
The decision tree node can be run in two modes: training or testing or simply used to visualize a tree from the general edit options dialog.
Full details of the options available for the discovery node can be found starting with the general options page of the decision tree wizard.
