基于遗传算法和分类树的信用分类方法

Credit classification based on genetic algorithm and classification tree

银行信贷信用评估本质上是个分类问题,已有统计和非统计的各种方法应用于信用评估,其中分类树方法,也称为递归分割法,比较适用于处理定性变量,而作为非统计方法之一的遗传算法则适用于处理连续型定量变量之间的非线性关系,但无法处理定性变量,利用这两种方法特点的互补性,构建了一种分类树和遗传算法相结合的信贷信用评估方法,先用分类树方法按照定性变量分类,然后在每个叶结点上用遗传算法按照定量变量分类.实证分析表明,该方法比单独使用分类树方法或遗传算法的分类准确率高.

The loan credit scoring is essentially a classification problem. tistical approaches have been applied in credit scoring. Among them, the is also called recursive partitioning approach is more suitable for dealing Various statistical and non-staclassification tree method which with qualitative variables than quantitative variables. Genetic algorithm as one of the non-statistical methods is suitable for treating the non-linear relationship of quantitative variables, but not good for treating qualitative variables. A new synthetic approach which combines the above two methods together for credit scoring is proposed. First, the classification tree method is applied for qualitative variables, then the genetic algorithm is applied at each terminal node for quantitative variables. Empirical test shows that this approach outperforms the genetic algorithm and classification tree method.