Explanation of attribute relevance in decision-tree induction

摘要

Strategist is an algorithm for strategic induction of decision trees in which attribute selection is based on the reasoning strategies used by doctors. The advantage is that in problem-solving applications of the induced decision tree, the relevance of an attribute or test can be explained in terms of the strategy it was selected to support, such as confirming a target outcome class or eliminating a competing outcome class. However, it is possible that an alternative approach to attribute selection may produce a decision tree with greater predictive accuracy from a given set of training data. The structure of the decision trees that an algorithm produces may also be an important factor in terms of problem-solving efficiency. We present a new algorithm for strategic induction of decision trees in which Strategist's multiple-strategy approach to attribute selection is replaced by the single strategy of increasing the probability of a target outcome class. While sharing Strategist's ability to explain the relevance of attributes in strategic terms, the new algorithm often produces more efficient decision trees than Strategist and matches the accuracy of ID3 on some data sets.