一种多值决策属性的决策系统的离散化算法

Method of Attributes Discretization for Multi-value Decision Attributes in Decision System

决策系统中连续属性的离散化,即实型属性空间向整型属性空间的映射,它是对决策表中属性约简的第一步.针对多值决策属性的决策信息系统,提出一种新的属性离散化算法.首先根据决策属性的不同,将条件属性集划分为不同的序列,对每两个序列求取候选断点,最后,综合所有的候选断点即为所求的候选断点集合;然后在基于条件属性重要度和贪心算法的基础上提出一种确定结果断点子集的新启发式算法.实例验证了本文所提出的算法能够取得较理想的连续属性离散化结果.

The discretization of continuous attributes values of a decision system which divides continuous values into different space and allocates some discrete values to each space is the fast step of attribute reducing for decision table. In this paper we propose one method of attributes discretization for multi-value decision attributes in decision System. First, we divide the condition attributes into different sequence according to decision attributes and calculate candidate cuts of every two sequence, then all the candidate cuts are unioned. At last, one heuristic method of caculating candidate cuts based on the importance of condition attributes and greedy algorithm is proposed. Moreover,excellent discretization results may be expected from them.