摘要
排序和分类是人类的两种基本的知识,一般文献中讨论的都是关于分类规则的挖掘,分类规则是一种刚提出的新思想,文章对犤3犦犤4犦提出的挖掘排序规则的算法作了更为全面深入的探讨和改进,所做的工作包括:比较可用于比较对象优劣的基于支配关系(dominancerelations)的扩充粗集理论的算法犤1,2犦和直接挖掘有序规则的算法犤3,4犦,分析各自的优缺点;指出犤3犦犤4犦中的算法存在两种情况下的对决策表的过分匹配;用定量翻译决策表的方法和翻译后的决策表的对称性改进犤3犦犤4犦的算法;针对翻译后的决策表基数往往很大和在扩充的粗集理论下是对联合(union)而不是对决策类求规则造成的时间复杂度高的问题,提出了一种启发式的寻找最小规则的算法。
Sorting and classifying are two basic kinds of knowledge.Extensive researches have been carried out on vari-ous aspects of generating classifying rules.Ordering rules is a noval thought.The algorithms for extracting classifying rules presented inare studied deeply and inproved in this paper.The contributes in the paper include:comparing the al-gorithm of inducing ordering rules directly and the algorithm based on extending the theory of rough sets by replacing the indiscernibility relation by the dominance relation,which both can be used to order objects and analyzing their ad-vantages and disadvantages respectively;pointing out that two kinds of overfitting existing in the algorithm studied in;improving the algorithm in translating the decision table quantitatively and reducing the size of the translated dicision table by exploiting its symmetry;inaugurating a heuristic algorithm to obtain an approximately minimal set of rules to counteract the high time complexity as an consequence of the high cardinality of the translated decision table and the rules are induced based on union instead decision class.
出处
《计算机工程与应用》
CSCD
北大核心
2004年第28期161-163,共3页
Computer Engineering and Applications
基金
国家自然科学基金项目(编号:60273043)
