Decision Trees

• A decision tree is a tree with a question at each node.
• The arcs are the answers.
• For categorisation, leaves are the final category.
• Decision trees are a good knowledge representation scheme. Apparently physicians are trained with this.
• C4.5 is a common algorithm for learning decision trees.
• Imagine all binary values and two categories.
• One way to build a tree is to see if any of the values split the categories perfectly. If so, make that the first question, and remove all the others from the training set. You can test this each time.
• Otherwise, use the feature that best splits the categories.
• Repeat as necessary.
• At the end, you can just guess.
• You can also do these things with more categories, larger numbers of input values and continuous input values.