有序决策树在SOCA下的扩展及模糊有序决策树的研究

发布时间：2018-01-17 19:33

本文关键词：有序决策树在SOCA下的扩展及模糊有序决策树的研究　出处：《河北大学》2015年硕士论文　论文类型：学位论文

【摘要】：简单有序分类方法(SOCA:Simple Ordinal Classification Approach)是Frank和Hall提出的通用方法,任何能给出样例概率估计的分类算法,如C4.5算法、K-近邻算法(KNN:k-Nearest Neighbor)和极限学习机(ELM:Extreme Learning Machine)算法等都能应用该方法来解决有序分类问题。但在SOCA中,只有决策属性的序信息被用于分类,而没有考虑条件属性的序信息。但是我们的实验研究发现条件属性的序信息能够改进分类算法的泛化能力。针对上述问题,本文将SOCA推广到有序决策树上,提出了一种改进的有序分类算法,该算法同时考虑了条件属性和决策属性的序信息。另外,本文还分析了SOCA对基本分类算法(如C4.5算法、K-近邻算法和ELM等)的敏感性。另外,我们还将有序决策树推广到了模糊环境,提出了一种模糊有序分类算法。并对本文提出的算法的性能进行了实验分析,实验结果显示本文提出的算法是行之有效的。
[Abstract]:Simple Ordinal Classification (SOCA). Is a general method proposed by Frank and Hall. Any classification algorithm, such as C4.5 algorithm, which can give sample probability estimation. KNN: k-nearest neighbor (KNN: k-nearest neighbor) and Ultimate Learning Machine (ELM: extreme Learning Machine). The algorithm can be used to solve the problem of ordered classification, but in SOCA. Only the order information of decision attributes is used for classification, but the order information of conditional attributes is not considered. However, our experimental research shows that the order information of conditional attributes can improve the generalization ability of classification algorithms. In this paper, SOCA is extended to an ordered decision tree, and an improved ordered classification algorithm is proposed, which takes into account the order information of both conditional attributes and decision attributes. In this paper, the sensitivity of SOCA to basic classification algorithms (such as C4.5 algorithm, K- nearest neighbor algorithm and ELM algorithm) is also analyzed. In addition, we extend the ordered decision tree to fuzzy environment. A fuzzy ordered classification algorithm is proposed, and the performance of the proposed algorithm is analyzed experimentally. The experimental results show that the proposed algorithm is effective.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：O225

【参考文献】