改进单调决策树算法-计算机应用技术专业论文.docxVIP

I I 摘 要 摘 要 在日常生活中，单调分类问题普遍存在。与一般的分类问题不同，单调分类要求分 类器分类时满足这样的单调约束：如果样例A的所有条件属性的值不小于样例B对应条 件属性的值，那么A的决策属性的值不小于B的决策属性的值。 本文分析了决策树算法在处理单调分类问题时存在的问题，针对其中的问题，本文 提出了改进的单调决策树算法，主要工作包括以下两个方面： （1）对连续值属性决策树归约，在选取割点时，应该选取能使决策树满足单调约束 的割点。在割点分割决策树后，只要保证左子树中所有样例的决策属性值不大于右子树 中的样例的决策属性值，就可以保证生成的决策树满足单调约束。基于这一思想并结合 优势粗糙集理论，本文提出了一种单调决策树算法。实验结果证明，该算法能保证生成 的决策树满足单调约束条件，与经典决策树相比，该算法有更好的分类精度。 （2）为了提高单调决策树算法的泛化能力，本文提出了基于矩阵的单调决策树扩展 模型，该模型能将本文中的单调决策树算法泛化到属性为离散值的情况。 关键词 单调决策树 单调分类问题 单调约束 优势粗糙集 II II Abstract Abstract The monotonic classification problem is commonly in real life. Different from the general classification problem, the monotonic classification classifiers should satisfy the monotonic constraint: if all values of the condition attributes of the sample A is not less than the ones of corresponding condition attributes of B, then the values of decision attribute A is not less than the ones of decision attribute B. By analyzing the problems of the monotonic classification problems, in order to deal with some of them, this paper proposed the improved monotonic decision tree algorithm. The main works include the following two aspects: （1）For the induction of decision tree for continuous-valued attributes, the selection of cut should ensure that the decision tree meet the monotonic constraint. After the cut point splitting the decision tree, as long as ensuring that the values of the decision attribute of the samples in the left branch is not more than the values of the decision attribute of the samples in the right branch, the decision tree can be guaranteed to meet the monotonic constraints. Based on this idea and the theory of dominance rough set, this paper proposes a monotonic decision tree algorithm. Experimental results show that the decision tree generated by our algorithm can meet the monotonic constraints. Compared with the classical decision tree, the proposed algorithm has higher classification accuracy. （2）In order to improve the generalization ability of the monoto