结合GBDT特征衍生与集成学习的客户忠诚度预测
Integrating GBDT Feature Derivation with Ensemble Learning for Customer Loyalty Prediction
龚安 1耿航1
作者信息
- 1. 中国石油大学(华东)青岛软件学院、计算机科学与技术学院 青岛 266580
- 折叠
摘要
为提升预测银行客户忠诚度的准确性,提出了结合GBDT特征衍生与集成学习的客户忠诚度预测方法.首先通过GBDT模型对原始数据集进行特征衍生,得到更多有区分度特征的新数据集.并在此基础上提出了以集成学习为核心的SV-XLC模型.模型总体上是对Stacking进行了改进.SV-XLC分为两个部分,初级预测模块和次级预测模块.初级模块由多个基础预测器组成,Voting模型内嵌在次级模块中,也由多个基础预测器组成.数据集经过五折交叉验证先传入初级模块的不同基础预测器中.经过处理后会产生新的数据集,将其代入次级模块的Voting的不同基础预测器训练预测,得到最终结果.论文在Kaggle的银行客户公开数据集上进行了验证.实验结果显示,该方法明显提高了银行客户忠诚度预测的准确性.
Abstract
In order to improve the accuracy of predicting bank customer loyalty,a customer loyalty prediction method combin-ing GBDT feature derivation and integrated learning is proposed.Firstly,the GBDT model is used to derive features from the origi-nal dataset to obtain a new dataset with more distinguishing features.And based on this,an SV-XLC model with integration learning as the core is proposed.The model is generally an improvement of Stacking.SV-XLC is divided into two parts,the primary predic-tion module and the secondary prediction module.The primary module consists of multiple base predictors,and the Voting model is embedded in the secondary module,which also consists of multiple base predictors.The dataset is first passed into the different base predictors of the primary module after a five-fold cross-validation.After processing,a new dataset is generated,which is substitut-ed into the different base predictors of Voting in the sub-module to train the prediction and get the final result.This paper is validat-ed on a public dataset of bank customers from Kaggle.The experimental results show that this method significantly improves the ac-curacy of bank customer loyalty prediction.
关键词
客户忠诚度预测/机器学习/GBDT/集成学习Key words
customer loyalty prediction/machine learning/GBDT/integrated learning引用本文复制引用
基金项目
中石油重大科技项目(ZD2019-183-004)
中央高校基本科研业务费专项资金(20CX05019A)
出版年
2024
国家科技期刊平台
NSTL
万方数据