基于构建有序决策树,提出了一种新的归纳算法。该算法选择的扩展属性不仅和类的有序互信息值最大,而且要求和同一分支上已被用过的条件属性的有序互信息值最小。实验结果表明,考虑了条件属性之间的相关性后,可避免同一条件属性的重复选择,真正体现了条件属性和决策属性之间的有序互信息,与已有的算法相比,提高了测试精度。 Based on the construction of an orderly decision tree, a new induction algorithm is proposed. The extended attribute chosen by the algorithm not only has the highest value of ordered mutual information with the class, but also requires that the ordered mutual information of the condition attributes that have been used on the same branch has the smallest value. The experimental results show that considering the correlation between conditional attributes, we can avoid the repeated selection of attributes with the same condition and truly reflect the orderly mutual information between conditional attributes and decision attributes. Compared with the existing algorithms, Test accuracy.