一种决策风险代价与属性偏好融合的适应性决策树算法

Adaptive Decision Tree Algorithm Based on Fusion of Decision Costs and Attribute Preference

针对现有决策树模型在分类过程中不能充分考虑决策精度、决策者的属性偏好以及决策风险因素的影响问题,提出一种决策风险代价与属性偏好融合的适应性决策树算法.算法结合决策粗糙集和代价敏感学习问题,引入用户偏好程度和决策风险损失函数的概念,根据贝叶斯最小风险决策原则,计算决策风险代价,通过构建适应度函数作为启发式函数选择划分属性,从而建立决策树模型.在决策树构建过程中,使用置信因子概念对决策树进行剪枝,以防生成的决策树过于庞大.实验结果表明该决策树算法是有效的,能充分考虑决策者的属性偏好和因决策的不确定性产生的误分代价,实验参数的设置可以增强算法的适应性,满足不同应用领域的需求.

In view of the fact that the existing decision tree models have not fully considered the decision accuracy,attribute preference behavior of decision makers and decision risk factors during the classification process,this paper proposes an adaptive decision tree algorithm based on fusion of decision costs and attribute preference. Combining decision-theoretic rough set and cost sensitive learning problems,by introducing the misclassification decision costs into the probabilistic approximations of the target,the model of decision-theoretic rough set is then sensitive to cost. Introducing the decision loss function and the concept of user preference degree because different users tend to have different preferences for each attribute in the decision table,and according to the Bayesian minimum-risk decision rules the decision costs are calculated. Through constructing adaptability-degree function as heuristic function is used to select test attributes,so as to establish a decision tree model. In the process of constructing the decision tree,the confidence factor is put forward to prune the decision tree and to prevent the generation of decision tree too large. The experimental results show that the decision tree algorithm is effective,which can fully take into account the decision maker＇s preference and the cost of misclassification due to the uncertainty of decision making,the experimental parameters can enhance the adaptability of the algorithm and meet the needs of different applications.