移动通信网用户流失预测

发布时间：2018-04-25 04:06

本文选题：用户流失预测 + 用户相关性　；参考：《中国科学技术大学》2016年硕士论文

【摘要】：近几年随着移动用户的急剧增长,移动通信市场接近饱和,移动用户流失已经成为运营商主要关心的问题。通过移动通信用户关系管理数据(CRM)和用户通话详单数据(CDR)来预测移动通信用户流失情况受到了运营商的广泛关注。现在不管是在理论研究还是实际应用,移动用户流失预测问题都没有得到很好的解决。因此进行移动用户流失预测研究具有重要的现实意义和广阔的研究前景。在移动通信大数据时代将要到来的背景下,本文使用移动用户数据进行用户流失预测分析工作。在研究内容上,本文进行了移动用户流失预测特征分析和移动用户流失预测建模两个方面的考察。在流失预测特征分析上,本文使用多种分析方法和技术从多角度重点分析用户语音业务相关性。利用修正阿兰方差(MAVAR)初步分析用户在同一基站通话到达的时间相关性,从而验证了用户语音业务行为的长时相关性及部分情况的短时相关性,并且通过分析实际通话到达过程与拟合的同均值泊松分布的卡方检验值,对于用户语音业务行为相关性进行深入分析验证了用户语音业务行为相关性。本文进一步从异质泊松过程的三大特点进一步分析用户语音业务行为是否符合异质泊松过程特点,并从分析结果中进一步发现用户语音业务行为不同时间尺度上的特点,进一步验证用户语音业务相关性。在流失预测分析中,本文使用后向传播人工神经网络(BPANN)作为移动通信用户流失预测的工具。根据流失预测特征分析中的结论,本文考虑用户相关性对用户流失行为的影响,因此选取的特征包括传统的用户静态信息特征、用户动态通话行为特征、用户相关性特征。并且根据运营商实际关心的用户流失预测问题,提出了建立分析用户45天内流失情况的预测模型,分析了加入用户相关性特征前后用户流失预测情况,发现加入用户相关性特征能大大提高流失预测准确度。通过初步的特征分析和定量特征重要性分析,进一步分析选取的特征对于流失预测的影响。本文得到的移动用户流失预测特征分析的相关结论可以提升人们对于移动通信网用户语音行为的认知,同时用户相关性分析结果可以应用于用户流失预测中。本文建立的用户流失预测模型对于运营商进行流失预测,采取用户挽留措施有重要指导意义。
[Abstract]:In recent years, with the rapid growth of mobile users, the mobile communication market is nearly saturated, the loss of mobile users has become the main concern of operators. Using CRM and CDR) to predict the loss of mobile communication users has been paid much attention by operators. At present, the problem of mobile user churn prediction has not been solved well either in theory or in practice. Therefore, the study of mobile user loss prediction has important practical significance and broad research prospects. Under the background of the coming of big data era, this paper uses mobile user data to predict and analyze the loss of users. In the aspect of research, this paper studies the feature analysis of mobile user churn prediction and the modeling of mobile user churn prediction. In the feature analysis of loss prediction, this paper uses a variety of analysis methods and techniques to analyze the relevance of user voice services from multiple perspectives. By using the modified Alain variance and MAVARA, the time correlation of the call arrival of the user at the same base station is analyzed preliminarily, which verifies the long time correlation of the user's voice service behavior and the short time correlation of some cases. By analyzing the actual call arrival process and the fitted Chi-square test value of Poisson distribution with the same mean value, the correlation of the voice service behavior of the user is analyzed in depth to verify the correlation of the user voice service behavior. In this paper, the characteristics of heterogeneous Poisson process are further analyzed, and the characteristics of different time scales are found from the analysis results. Further verify the relevance of user voice service. In the analysis of loss prediction, BPANN (back propagation artificial neural network) is used as a tool to predict the loss of mobile communication users. According to the conclusion of loss prediction feature analysis, this paper considers the influence of user correlation on user churn behavior, so the selected features include traditional static information feature, dynamic call behavior feature, and user correlation feature. According to the problem of user churn prediction that operators are concerned about, this paper puts forward a forecast model to analyze the loss of users within 45 days, and analyzes the forecast of user loss before and after adding the characteristics of user correlation. It is found that adding user correlation features can greatly improve the loss prediction accuracy. The influence of selected features on loss prediction is further analyzed by means of preliminary feature analysis and quantitative feature importance analysis. The results obtained in this paper can improve people's cognition about the voice behavior of mobile communication network users, and the results of user correlation analysis can be applied to user churn prediction. The prediction model of user loss established in this paper is of great significance for operators to predict the loss and to take measures to retain users.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN929.5

【相似文献】