【作者】 李莉；

【导师】 刘继成；

【作者基本信息】 华中科技大学 ， 统计学， 2019， 硕士

【摘要】 数据挖掘是一种新型的数据分析技术.数据挖掘中的算法按照其应用不同可以分为监督学习:神经网络、支持向量机、决策树和回归等;无监督学习:聚类分析、关联规则分析;数据降维:主成分分析、因子分析等.常见的决策树算法包括:ID3算法、C4.5算法、CART算法,其中ID3算法作为最早的决策树算法,应用最为广泛.本文针对决策树中ID3算法进行深入研究,解决了该算法中存在的多值偏向、信息熵中对数运算较为复杂的问题.针对ID3算法存在的多值偏向问题,本文提出了一种修改信息增益函数的ID3算法,将属性与类别之间的相关系数、属性值的个数引入到信息增益函数中.改进的ID3算法减小了属性值较多且与类别相关性不大属性的信息增益值,解决了多值偏向问题;针对信息熵中复杂的对数运算,本文利用泰勒公式简化了信息熵公式,将信息熵公式中的对数运算转换为非对数运算;通过UCI上四个经典的数据集,数值检验了改进后的ID3算法的确提高了分类精确度,简化的信息熵公式降低了时间复杂度.最后,本文将改进的ID3算法应用于银行信用卡违约评估中,对其数据集中的诸如缺少属性值、属性值离散化和属性选择等问题提出了具体的解决方案.相比之下,改进的ID3算法提高了分类精确度并降低了算法复杂度,该实例也为银行工作人员提供了决策支持.更多还原

【Abstract】 Data mining is a new data analysis technology.Algorithms in data mining can be divided into supervised learning: neural network,support vector machine,decision tree and regression,unsupervised learning: clustering analysis,association rule analysis,data dimensionality reduction: principal component analysis,factor analysis and so on.Common decision tree algorithms include ID3 algorithm,C4.5 algorithm and CART algorithm.ID3 algorithm,as the earliest decision tree algorithm,is widely used.In this paper,ID3 algorithm in decision tree is studied in depth,and the problems of multi-value bias and logarithmic operation of information entropy in this algorithm are solved.Aiming at the multi-value bias problem of ID3 algorithm,this paper proposes a modified ID3 algorithm of information gain function,the correlation coefficient between attributes and categories,the number of the attribute value is introduced into the information gain function.The improved ID3 algorithm reduces the information gain value of attributes with more attributes and little correlation with categories,and solves the problem of multi-value bias.In view of the complex logarithmic operation in information entropy,this paper simplifies the information entropy formula by using Taylor formula,and transforms the logarithmic operation in information entropy formula into non-logarithmic operation.The improvement is verified numerically through four classical data sets on UCI.The ID3 algorithm improves the classification accuracy and reduces the time complexity by simplifying the information entropy formula.Finally,this paper applies the improved ID3 algorithm to the credit card default assessment of banks,and puts forward specific solutions to the problems such as lack of attribute values,discretization of attribute values and attribute selection in its data set.In contrast,the improved ID3 algorithm improves the classification accuracy and reduces the complexity of the algorithm.This example also provides decision support for bank staff.更多还原