【摘要】 连续属性的决策表知识获取有两个问题需要解决 :其一是连续属性的离散化问题。这个问题已引起了人们的注意 ,在简单评述的基础上 ,结合粗集理论 ,提出了一种新的数据离散化方法。大致思想是先用 K-W检验方法粗略评价各连续属性的重要性 ,然后用 c均值聚类给出各属性的量化结果。若决策表不相容 ,则按属性重要性依次增加分类的区间数 ,如此反复直到决策表相容为止。其次是最优属性的选择问题。在此借助于简约格和模拟退火算法 ,给出一种启发式算法和最优算法。实例表明 ,上述方法是有效的。更多还原

【Abstract】 It is a difficult task to determine the optimal attribute set in decision tables containing continuous attributes. Two problems have to be solved. The first is the discretization of continuous attributes which has been discussed widely. In the paper, some discritization methods are overviewed and a new discretization one is put forward. The method first utilizes K W test to value the relative importance of every continuous attrubute to decision attributes roughly. Then discretized values are given using c mean value clustering. If in consistency appears in the decision table, then the values should be increased one by one according to the inportance of attributes until there is consistency in the decision table. The second problem is the choice of optimal attribute set. In this part, we first propose a heuristic algorithm using reduct lattice in rough set theory but its efficiency is low. To do so, the simulated annealing algorithm is used. The examples show that it is better.更多还原