In the custom credit analysis, decision tree is always used to separate the custom because of the large amount of nominal variables. Achieving the global best tree is an NP-hard problem, so greedy method is the most common and fast way to get a relative good solution yet seldom the global best one. The new method in this paper provides a brand new idea to breed the decision tree using genetic algorithm to get a relative good tree during the acceptable time. And then, combined with logistic regression and multi-layer perception, a credit analysis system will be generated. |