智能化调参的XGBOOST算法及其在电信营销中的应用  

Intelligent Parameter Adjustment XGBOOST and Its Application in Telecom Marketing

作　　者：成晨 程新洲[1] 张恒[1] 韩玉辉[1] Cheng Chen;Cheng Xinzhou;Zhang Heng;Han Yuhui(China Unicom Network Technology Research Institute,Beijing 100048,China)

机构地区：[1]中国联通网络技术研究院,北京100048

出　　处：《邮电设计技术》2018年第10期20-24,共5页Designing Techniques of Posts and Telecommunications

摘　　要：XGBOOST是一种集成学习算法,通常采用网格法调参,参数优化幅度有限;提出连续量子粒子群算法,并将其引入XGBOOST的调参过程,得到全局优化参数,从而提升XGBOOST算法性能;同时将该算法应用于电信精准营销案例——运营商用户换机预测场景中,得到了基于连续量子粒子群算法调参的XGBOOST用户换机预测模型,与基于网格法调参的XGBOOST相比,该模型获得了更高的用户换机的预测准确度。XGBOOST is an ensemble learning algorithm,which usually uses grid method to adjust parameters and the parameter optimi- zation scope is limited. The algorithm of parameters adjusting based on Continuous Quantum-inspired Particle Swarm Optimi- zation is proposed,which improves the performance of XGBOOST. This model has been applied in a case of telecom precision marketing,the prediction of user equipment replacing.Compared with XGBOOST based on Grid parameter adjustment method, the model achieves higher prediction accuracy of user equipment replacing.

关 键 词：连续量子粒子群 XGBOOST 机器学习 电信精准营销 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]