Alternative Evaluation Criterion of Constructing Decision Tree of Discrimination

An alternative criterion of evaluating the built decision tree for discrimination is discussed in this paper. The maximal decrease in undistinguished degree is proposed as the principle to choice explanatory variable for building a binary tree. From the basic concept, this new criterion is approximately consistent with the classical method using decrease in impurity. Though these two criteria are equivalent if choosing appropriate weights for the classical algorithm of decrease in impurity. The research shows that the computation of new criterion is more simple. Moreover, the result from the case study shows that the new criterion is more interpretable.