The algorithm used to create trees is very simple. It can be summarised in four steps:
For each cue, select a classification threshold that maximizes the difference between the hit-rate (HR) and false alarm rate (FAR) of classifications of all data based on that cue. If the cue is numeric, the threshold is a number. If the cue is a factor, the threshold is one or more factor levels. These thresholds are calculated completely independently of all other cues.
Rank cues in order of their highest HR - FAR value calculated in step 1.
Create all possible trees by varying the exit point at each level to a maximum of 5 levels.
Reduce the size of trees by removing lower levels containing less than 5% of the dataset.