论文部分内容阅读
本项目通过对用户的移动业务使用行为进行分析,锁定用户离网的一系列关键影响因素,通过数学模型的搭建及优化,建立起行之有效的离网预测模型。本项目对选取XX电信移动业务用户作为研究对象,结合数据源的信息,将移动用户离网预测提取为二分类场景,将次月的离网标签作为分类模型训练标签。主要选取三种模型方案:Logistic Regression分类算法、Random Forest分类算法、Gradient Boosting分类算法,进行分类训练和预测。最终的预测结果是通过分类模型输出用户离网的参考概率,以参考概率作为排序依据对用户的离网倾向进行排序。
This project analyzes a series of key influencing factors of users’ off-grid by analyzing user’s mobile service usage. Through establishing and optimizing mathematical models, this project establishes an effective off-grid forecasting model. In this project, XX telecom operators are selected as research subjects. Based on the data source information, mobile users’ off-grid prediction is extracted into two categories and the next month’s off-grid labels are used as training models for classification models. Three models are mainly selected: Logistic Regression classification algorithm, Random Forest classification algorithm, Gradient Boosting classification algorithm for classification training and prediction. The final prediction result is that the reference probability of the user leaving the network is output through the classification model, and the user’s offline tendency is sorted based on the reference probability.