基于数据挖掘的电信客户流失研究

发布时间：2018-03-25 18:22

本文选题：支持向量机　切入点：决策树　出处：《淮北师范大学》2013年硕士论文

【摘要】：随着电信行业竞争的加剧,客户流失分析与预测已经成为客户关系管理的重要内容。电信客户行为数据的特征呈现出高维度、数据偏斜、非线性。传统的方法难以消除数据之间的冗余以及找到线性规律,使得预测正确率较低。同时开发系统过程繁杂,收益偏低,挽留客户成本过高。本文从以解决以上问题为出发点,主要研究基于支持向量机的客户流失预警模型。支持向量机的算法复杂度随着样本数据维度和样本总数量的增加成几何数增长。针对这个问题,提出了一种改进的支持向量机分类方法。通过引入分类圆心、分类半径、分类圆心距等概念,从而更加快速准确的删除非支持向量点,引入混淆度的概念,解决了如何在样本严重混淆的时候进行剔除混淆点,保证算法的泛化性。实验证明,采用这种改进的算法能够在严重混淆的训练样本中保证准确度的同时提高支持向量机分类速度。我们在Clementine数据挖掘工具平台的基础上设计了基于传统支持向量机、改进支持向量机、决策树和神经网络的客户流失预警模型。根据实验结果对各种分类算法进行了比较,得出了一个针对样本数据的客户流失原因报告。通过本文的研究,我们解决了客户流失预警系统开发费用高,预测效率低下,预测正确率不高的问题。设计了客户预警流失模型,为企业制定挽留客户决策提供了技术支撑。
[Abstract]:With the increasing competition in telecommunication industry, customer churn analysis and prediction has become an important part of customer relationship management. The characteristics of telecom customer behavior data show a high dimension and data skew. The traditional method is difficult to eliminate the redundancy between data and find the linear rule, which makes the prediction accuracy low. At the same time, the process of developing the system is complicated, the income is low, and the cost of retaining customers is too high. From the point of view of solving the above problems, This paper mainly studies the customer churn warning model based on support vector machine (SVM). The complexity of SVM algorithm increases with the increase of the dimension of sample data and the total number of samples. In this paper, an improved classification method of support vector machine is proposed. By introducing the concepts of the classification center, the classification radius and the classification center distance, we can delete the non-support vector points more quickly and accurately, and introduce the concept of the degree of confusion. How to eliminate the confusion points when the samples are seriously confused is solved, and the generalization of the algorithm is ensured. The improved algorithm can improve the classification speed of support vector machines while ensuring accuracy in seriously confused training samples. On the basis of Clementine data mining tool platform, we design a customer churn early warning model based on traditional support vector machine, improved support vector machine, decision tree and neural network. A customer churn report for sample data is obtained. Through the research of this paper, we solve the problems of high development cost, low prediction efficiency and low prediction accuracy of customer churn early warning system, and design customer early warning loss model, which provides technical support for enterprises to make customer retention decision.
【学位授予单位】：淮北师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP311.13;TP181

【参考文献】