| 摘要: |
| 准确地识别有出境意向的用户具有重要的意义，可为出境服务企业的精准营销实施、出境运营的高效管理和政策制定提供决策支持。针对此需求，提出了一种基于多分类器集成和特征融合的用户出境预测方法。首先利用用户的移动终端信息交互数据，挖掘用户的出境相关行为特征和静态特征作为样本特征；其次，通过最小冗余最大相关算法筛选最优特征，并利用贝叶斯优化算法寻找多个分类器最优超参数；最后，基于集成学习思想构建三层架构的用户出境预测模型，模型通过融合前两层分类器的输出特征生成新特征，并将其输入第三层分类器进行学习和预测。实验表明，所提方法的F1值和AUC(Area under the Curve)值分别达到了97.16%和97.21%，对于用户出境具有较高的预测精度。 |
| 关键词: 数据挖掘 移动大数据 用户出境预测 三层融合模型 特征融合 |
| DOI： |
|
| 基金项目:教育部-中国移动科研基金项目(MCM20170203)；重庆市技术创新与应用示范(产业类重点研发)项目(cstc2018jszx-cyzdX0124)；重庆邮电大学人才引进项目(A2017-10) |
|
| Users’ outbound prediction based on ensemble learning and features fusion |
| ZHANG Xuan,XU Guoliang,WEI An,WANG Chao,LUO Jiangtao |
| (a.School of Communication and Information Engineering;b.Electronic Information and Networking Research Institute,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| It is of great significance to accurately identify mobile phone users who will go abroad in the near future.It can provide decision support for the precise marketing implementation of outbound service companies,the efficient management of outbound operations and policy formulation.Based onensemble learning and features fusion,a method for mobile phone users′ outbound prediction is proposed.The method first uses the information interaction data of the user′s mobile terminal,and mines users′ specific behavioral features and attributes features as sample features.Secondly,the optimal features are selected by using the algorithm of minimum redundancy maximum correlation,and multiple classifiers are adjusted based on Bayesian optimization algorithm.Finally,a three-level model for users′ outbound prediction is constructed by using ensemble learning method.The model generates new features by fusing the output features of 1-level and 2-level classifiers,and then uses the 3-level classifier to train and predict new features.The experimental results show that the F1-score and the area under the curve(AUC) value of the proposed model is 97.16% and 97.21%,respectively,and a higher prediction accuracy for users′ outbound is obtained. |
| Key words: data mining mobile big data users′ outbound prediction three-level fusion model feature fusion |
