客户流失预测模型研究及其应用

发布时间：2018-05-09 03:40

本文选题：客户流失 + 预测　；参考：《西北大学》2017年硕士论文

【摘要】：随着全球的商业竞争愈来愈激烈,客户流失预测已经成为客户关系管理中非常重要的内容。预测即将流失的客户,并制定相应的措施挽留客户已经成为促进企业发展的关键性因素。本文从对电信和信用卡客户的行为数据分析入手,针对其中的冗余特征和正负类样本不均衡等特点,提出一种新的特征选择算法和非均衡数据处理算法,以此建立一种新的客户流失预测模型。本文的主要工作内容如下:1.针对客户信息数据量大且特征冗余的问题,提出一种多指标融合的mRMR特征选择算法(MIF-mRMR)。该算法对mRMR算法进行改进,将马氏距离和最大信息系数相结合来综合评定特征和类别、特征和特征之间的相关性。实验结果证明,该算法所选择的特征子集维度较小,且与mRMR算法相比准确率提高3%左右。2.针对客户数据中流失客户与正常客户比例相差较大的问题,提出一种基于边界混合采样的非均衡数据处理算法(BMS)。该方法首先通过引进“变异系数”找出样本的边界域和非边界域,然后对边界域中的少数类样本进行过采样处理,非边界域中的多数类样本进行欠采样处理,从而实现原始样本的均衡化。实验结果证明,该算法与其它3种流行的非均衡数据处理算法相比均能有效地提升少数类样本的识别性能。3.基于以上内容,提出一种新的客户流失预测模型。该模型首先分别使用MIF-mRMR算法和BMS算法对数据进行特征选择和均衡化,然后将均衡化后的数据送入SVM、C4.5和随机森林3种分类器进行流失客户的预测。实验结果表明,使用该模型能够得到较好的客户流失预测结果,同时SVM分类器相较其他两种分类器更适合客户流失预测的研究。
[Abstract]:With the global business competition becoming more and more fierce, customer churn prediction has become a very important part of customer relationship management. It is a key factor to promote the development of enterprises to predict the customers that will soon be lost and to formulate corresponding measures to retain them. Based on the analysis of behavior data of telecom and credit card customers, a new feature selection algorithm and unbalanced data processing algorithm are proposed in this paper. Based on this, a new customer churn prediction model is established. The main work of this paper is as follows: 1. Aiming at the problem of large amount of data and redundant features of customer information, a multi-index fusion mRMR feature selection algorithm (MIF-mRMRA) is proposed. The algorithm improves the mRMR algorithm and combines Markov distance with the maximum information coefficient to evaluate the correlation between features and categories, features and features. Experimental results show that the feature subset dimension chosen by the algorithm is small and the accuracy of the proposed algorithm is about 3% higher than that of mRMR algorithm. In order to solve the problem that the ratio of customer loss and normal customer is different from that of normal customer, a non-equilibrium data processing algorithm based on boundary mixed sampling is proposed. First, the "coefficient of variation" is introduced to find out the boundary and non-boundary regions of the samples, then a few kinds of samples in the boundary domain are oversampled, and most of the samples in the non-boundary domain are under-sampled. In order to realize the equalization of the original samples. Experimental results show that the proposed algorithm can effectively improve the recognition performance of a small number of samples compared with the other three popular unbalanced data processing algorithms. Based on the above, a new customer churn prediction model is proposed. The model firstly uses MIF-mRMR algorithm and BMS algorithm to select and equalize the data, and then the equalized data are sent into SVMC4.5 and random forest classifier to predict the loss of customers. The experimental results show that the model can get better customer churn prediction results, and the SVM classifier is more suitable for customer churn prediction than the other two classifiers.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;TP301.6

【相似文献】