一种基于双层融合结构的客户流失预测模型  

Customer Churn Prediction Model Based on Double Layer Fusion Structure

作　　者：李为康 杨小兵 LI Wei-kang;YANG Xiao-bing(College of Information Engineering,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学信息工程学院,杭州310018

出　　处：《小型微型计算机系统》2020年第8期1634-1640,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61303146)资助。

摘　　要：针对客户流失预测精准性的提升,提出了一种基于双层融合结构的客户流失预测模型.该模型不需要提前对数据集进行独热编码,避免了维度灾难和数据稀疏问题.其主要思想是融合多个高准确率的基于树的机器学习算法组成一个包含Stacking层与Voting层的双层预测模型.数据集经过处理后输入到Stacking层,然后Stacking层的预测结果与处理后的数据集合并传递给Voting层,同时将Stacking层加入到Voting层的预测中,最后输出最终的预测结果.在Kaggle的电信客户公开数据集上的实验表明,与经典的客户流失预测模型和改进的客户流失预测模型相比,本模型明显提高了客户流失预测的准确率和精准率.Aiming at the improvement of customer churn prediction accuracy,a customer churn prediction model based on double-layer fusion structure is proposed.The model does not need One-Hot encoding in advance for the data set,avoiding dimensional disasters and data sparseness.The main idea of the model is to combine multiple high-accuracy tree-based machine learning algorithms to form a double-layer prediction model including the Stacking layer and the Voting layer.The data set is processed and input to the Stacking layer,and then the prediction result of the Stacking layer and the processed data set are transmitted to the Voting layer,and the Stacking layer is added to the prediction of the Voting layer,finally the final prediction result is output.Experiments on Kaggle’s telecom customer open dataset show that this model significantly improves the accuracy and precision of customer churn predictions compared to the classic churn prediction model and the improved customer churn prediction model.

关 键 词：客户流失预测 准确率 机器学习 分类模型 精准率 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]