基于随机KNN特征选择的高质量移动通信用户预测  

Prediction for High-Value Mobile Users Based on Random KNN Feature Selection

作　　者：崔伟[1] 夏汛[1] 孙瑜鲁 

机构地区：[1]泸州职业技术学院,泸州646000 [2]四川大学电子信息学院,成都610064

出　　处：《现代计算机（中旬刊）》2017年第9期9-12,共4页Modern Computer

基　　金：川大-泸州战略合作科技项目(No.2015CDLZ-S12)

摘　　要：高价值移动通信用户预测是电信企业客户管理的一项重要内容,针对用户数据维度较高,规模较大,类不平衡较严重等问题,提出一种基于随机KNN的特征选择的预测方法,首先对初始数据进行随机采样构建多个KNN分类器,随后计算特征的权重以评估其重要性,利用广义顺序后退法对特征进行选择获得最优的特征子集,最后在结合集成学习方法中加入加权投票机制,建立预测模型。实验结果表明,该预测模型能够有效降低样本特征维度并提升对高价值移动通信用户预测性能。The prediction for high value mobile communication user plays an important role in the telecom enterprise customer management. Aimingat the problems such as high user data dimension, large scale and serious unbalanced class, proposes a method of feature selection basedon random KNN. Firstly, the initial data is randomly sampled to construct multiple KNN classifiers, and then the weights of the features arecomputed to measure its importance, and the generalized sequential backward selection method is used to select the optimal features sub-set. Finally, the weighted voting mechanism is added in the ensemble learning method to establish a predictive model. The experimental re-sults show that the model can effectively reduce the dimensions of the sample features and improve the prediction performance of the highvalue mobile communication users.

关 键 词：不平衡数据集 特征选择 K近邻 预测模型 

分 类 号：TP393.4[自动化与计算机技术—计算机应用技术]