基于半监督学习的客户流失预测及其在物流企业中的应用研究

发布时间：2018-02-25 09:19

本文关键词： 物流企业客户流失预测半监督学习协同训练非均衡数据分类　出处：《合肥工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着物流行业竞争的加剧,各物流企业提供的服务质量越来越高,且差异越来越小,客户的选择也越来越多,所以客户流失是每个物流企业都不得不面对的难题。由于客户流失会给物流企业带来巨大的损失,因此,客户流失问题受到了企业界和学术界的广泛关注。目前,已有大量的客户流失预测方法来解决该问题,主要分为传统分类方法和非均衡数据分类方法。虽然这两种方法已经取得了较好的预测效果,但这些方法往往需要大量的有标记客户数据。而在物流企业的客户流失预测中,获取大量的有标记客户数据代价昂贵。因此,如何综合利用有标记客户数据和无标记客户数据来对物流企业的客户流失进行预测,成为当前亟待解决的问题。半监督学习是一种能够有效利用有标记数据和无标记数据进行学习的方法。因此,本研究将半监督学习中的Co-training方法应用到物流企业客户流失预测问题中,来解决其存在的数据非均衡分布和获取有标记客户数据代价昂贵的问题。首先,本研究系统分析了客户流失预测和半监督学习的研究现状,分析了目前客户流失预测和半监督学习研究中存在的问题。其次,本研究对客户流失预测、物流企业中的客户流失预测和数据挖掘的基础理论进行了系统研究,分析了客户关系管理概述、客户流失概念及类型和客户流失预测过程,以及物流企业客户关系管理的特点、物流企业客户流失的主要原因和物流企业客户流失预测的两类错误,还分析了数据挖掘概述和数据挖掘中的两类分类方法。然后,在此基础上,针对客户流失预测中存在的数据非均衡分布和半监督学习问题,本研究将Co-training方法应用到客户流失预测中,并构建了基于半监督学习的客户流失预测模型。最后,以物流企业为应用背景,基于本研究提出的模型,本研究开发了面向物流企业的客户流失预测原型系统。通过将本研究提出的客户流失预测模型应用到实际应用场景中,来对模型的有效性和实用性进行验证。实验结果表明,本研究提出的模型在实际应用中取得了较好的预测效果。通过本研究,一方面,从客户流失预测研究中存在的问题入手,将Co-training方法应用到客户流失预测问题中,构建了基于半监督学习的客户流失预测模型,丰富和完善了客户流失预测的理论研究体系;另一方面,将提出的客户流失预测模型应用到物流企业中,开发了面向物流企业的客户流失预测原型系统,为准确地预测物流企业的客户流失提供了行之有效的方法。
[Abstract]:With the logistics industry competition, service quality and the logistics enterprise is higher, and the difference is more and more small, customers choose more and more, so the loss of customers is a problem that every logistics enterprises have to face. Because of the loss of customers will bring huge losses to the logistics enterprises so the customer churn problem has been widespread business circles and academic circles. At present, there are a large number of customer churn prediction method to solve this problem, mainly divided into the traditional classification methods and imbalanced data classification methods. Although these two methods have achieved good prediction results, but these methods often require a large amount of labeled data. And the customer in logistics enterprises the customer churn prediction, access to a large number of labeled customer data is expensive. Therefore, how to use customer data marked and unmarked customer data to logistics The loss prediction of enterprise customers, has become an urgent problem to solve. Semi supervised learning is an effective use of labeled data and unlabeled data to learn the method. Therefore, this study will prediction problem of semi supervised Co-training learning method applied to the logistics enterprise customer churn, to solve the existing data balanced distribution and access to customer data marked expensive problems. Firstly, this study analyzes the research status of customer churn prediction and semi supervised learning, analyzes the problems of customer churn prediction and semi supervised learning problems in the research. Secondly, the research of customer churn prediction, the basic theory of logistics enterprise customer churn prediction and data mining is studied and analyzed CRM, customer churn prediction process and loss of the concept and types of customers, and Flow characteristics of enterprise customer relationship management, two kinds of error causes of customer churn prediction of logistics enterprise and logistics enterprise customer churn, analyzes two classification methods overview of data mining and data mining. Then, on this basis, according to the customer churn prediction in the presence of data distribution and non balanced semi supervised learning problems in this study, the Co-training method is applied to the customer churn prediction, and the construction of the semi supervised learning based customer churn prediction model. Finally, taking the logistics enterprise as the background, this study based on the proposed model, this study developed for logistics enterprise customer churn prediction system. The forecast model will be applied to practical application in the scene the loss to the customer, the validity and practicability of the model is verified. The experimental results show that the application of the model proposed by this study in practice Get a better prediction result. Through this research, on the one hand, the customer churn prediction research in the existing problems, the Co-training method is applied to the prediction of customer churn, based on semi supervised learning model of customer churn prediction, enrich and improve the customer churn prediction theory research system; on the other hand, the customer churn prediction model is applied to the logistics enterprise, development oriented logistics enterprise customer churn prediction prototype system provides effective method for accurate prediction of customer churn of logistics enterprises.

【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274;F253

【相似文献】