Studies on Reducing the Necessary Data Size for Rule Induction from the Decision Table by STRIM
Studies on Reducing the Necessary Data Size for Rule Induction from the Decision Table by STRIM
复制标题
STRIM 减少决策表规则归纳所需数据量的研究
DOI：
10.1007/978-3-030-22815-6_11
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
T. Saeki
中科院分区：
文献类型：
--
作者：
Y. Kato;T. Saeki
提出了一种从决策表中归纳if-then规则的统计测试规则归纳方法。CNOM通过基于该表的统计检验来判断尝试规则的显著性。尝试规则的判断方法是基于标准正态分布对决策属性值的分布进行近似，因此判断方法需要满足近似条件的适当大小的数据集。本文提出了一种新的检验方法-minor-minor-M，它不采用近似分布检验，而采用原始分布检验，从而将其适用范围扩大到不满足条件的情形。具体地说，minor-CARM使用二项分布进行测试，并通过使用模拟实验与传统CARM相比，显示了适用范围的扩展和性能评估。仿真还表明，它给出了讨论和确认信息的有效性，从应用minor-CARM到现实世界的数据集所获得的结果。
STRIM (Statistical Test Rule Induction Method) has been proposed for an if-then rule induction method from the decision table. STRIM judges the significance of a trying rule by a statistical test based on the table. The method judging the trying rule has been executed based on the standard normal distribution approximating the distribution of the decision attribute’s values so that the judging method needs the proper size dataset satisfying the conditions of the approximation. This paper proposes a new STRIM named minor-STRIM not incorporating the test by the approximating distribution but by the original distribution, which expands the applicable range to cases not satisfying the conditions. Specifically, minor-STRIM uses a binomial distribution for the testing and shows the applicable range expanded and performance evaluation by use of a simulation experiment compared with those by the conventional STRIM. The simulation also shows that it gives discussing and confirming information the validity of the results obtained from applying minor-STRIM to a real-world dataset.