摘要
由于神经网络模型缺乏透明性,通过神经网络获得的知识难以被用户所理解,因而限制了它的发展。通过数据分类可以为神经网络提供一个解释机制,用规则来取代权值矩阵,可以较好的解决神经网络的“黑箱”问题。本文通过对分解式和示范式各种数据分类算法的分析,概括了它们的基本思想并对各种算法的性能进行比较,为实际应用领域不同类型和不同层次信息的数据分类选择提供决策说明。
Neural networks’ development is restricted because of its lack of transparent and its difficulty of understanding. A expound method can be provided by classification of data and rule extraction can replace weight matrix. Thus, the problem of black boxes of neural networks is resolved. This paper discusses the signification and feature of the classification based on artificial neural networks for data mining, the author analysis a lot of algorithms. The algorithms of the decomposition and pedagogical rule extraction are compared in detail, policy illustration is provided for choose of classification of data of different level and different type to real application domain.
出处
《辽宁工程技术大学学报（自然科学版）》
CAS
北大核心
2004年第4期507-509,共3页
Journal of Liaoning Technical University (Natural Science)
基金
国家自然科学基金项目资助(50207004)
关键词
神经网络
数据分类
规则提取
分解式
示范式
neural networks
classification
rule extraction
decomposition
pedagogical
