基于神经网络的模糊决策树改进算法  

Improved Algorithm of Fuzzy Decision Tree Based on Neural Network

作　　者：张敏[1] 彭红伟 颜晓玲 ZHANG Min;PENG Hongwei;YAN Xiaoling(College of Information Engineering,Dalian University,Dalian,Liaoning 116622,China)

机构地区：[1]大连大学信息工程学院,辽宁大连116622

出　　处：《计算机工程与应用》2021年第21期174-179,共6页Computer Engineering and Applications

基　　金：辽宁省重点研发计划(2017104014);辽宁省自然科学基金面上项目(20180551247)。

摘　　要：传统决策树通过对特征空间的递归划分寻找决策边界,给出特征空间的“硬”划分。但对于处理大数据和复杂模式问题时,这种精确决策边界降低了决策树的泛化能力。为了让决策树算法获得对不精确知识的自动获取,把模糊理论引进了决策树,并在建树过程中,引入神经网络作为决策树叶节点,提出了一种基于神经网络的模糊决策树改进算法。在神经网络模糊决策树中,分类器学习包含两个阶段:第一阶段采用不确定性降低的启发式算法对大数据进行划分,直到节点划分能力低于真实度阈值ε停止模糊决策树的增长;第二阶段对该模糊决策树叶节点利用神经网络做具有泛化能力的分类。实验结果表明,相较于传统的分类学习算法,该算法准确率高,对识别大数据和复杂模式的分类问题能够通过结构自适应确定决策树规模。Traditional decision tree finds the decision boundary by recursively dividing the feature space.It gives a“hard”division of the feature space and accurately describes of data.However,when dealing with big data and complex model problems,this precise decision boundary reduces generalization ability of decision tree.In order decision tree automatically acquire inaccurate knowledge,fuzzy theory is introduced into decision tree and in the process of building,neural network is introduced as decision leaf node.An improved algorithm of fuzzy decision tree based on neural network is proposed.In neural network fuzzy decision tree,classifier learning consists two stage:the first stage uses a heuristic algorithm with reduced uncertainty to divide the big data,and stops the growth of fuzzy decision tree until the node dividing ability is below reality threshold.The second stage uses neural network to classify the leaf node with generalization ability.Experiments show that compared with traditional classification learning algorithm,proposed algorithm has higher accuracy,and can determine the size of decision tree through structural adaptation for classification problem of big data and complex patterns.

关 键 词：决策树 神经网络 模糊理论 神经网络模糊决策树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]