基于神经网络的分类决策树构造

目前基于符号处理的方法是解决分类规则提取问题的主要方法,而基于神经网络的连接主义方法则用的不多,其主要原因在于虽然神经网络的分类精度高,但难于提取其所隐含的分类规则与知识.针对这个问题,结合神经网络的具体特点,该文提出了一种基于神经网络的构造分类决策树的新方法.该方法通过神经网络训练建立各属性与分类结果之间的关系,进而通过提取各属性与分类结果之间的导数关系来建立分类决策树.给出了具体的决策树构造算法.同时为了提高神经网络所隐含关系的提取效果,提出了关系强化约束的概念并建立了具体的模型.实际应用结果证明了算法的有效性.

A New Approach to Establish Decision Trees for Data Mining Based on Neural Network Learning

： The symbolic approaches currently are the main methods for classification rules mining．But connectionist approaches based on neural networks are not used frequently．The major reason is that although neural networks can give a high precise in classification,it is difficult to acquire the rules and knowledge generated by neural networks．To this problem,this paper presents a new approach to establish decision trees based on neural networks．The main idea of the approach presented is as follows：firstly through neural network learning, the relation model of attributes and classification of the data mining problem is established,then the decision tree is constructed based on analysis the derivative relations of classification and those attributes．The detailed algorithms for establishment of decision tree are presented．To improve the effectiveness of analysis of implicit relations generated by neural networks,the conception and concrete models of relation－enhancing constraints are also presented． The results of practice have proved the effectiveness of this approach．