基于改进的Tomek Link方法预测电信客户流失

Predicting Telecom Customer Churn Based on Improved Tomek Link Method

为提高电信客户流失检测的准确性和效率,提出了一综合方法。首先应用XGB-RFE方法对特征进行筛选,以选择最相关的特征,其次采用基于分层交叉验证框架下的数据平衡技术,来处理不平衡数据。通过这两种方法的综合应用,旨在提高模型性能和可解释性。结果表明,基于分层交叉验证框架下的Tomek Link欠采样技术,显著提高了各个模型的性能。此外,将该方法应用于TabNet模型中,同样取得了良好的效果。这一综合方法对于电信客户流失预测具有实际应用价值,有望提高流失检测率,改善业务决策。

In order to address the issue of low detection rates in telecom customer churn,a comprehensive approach is proposed.Firstly,the XGB-RFE method is applied for feature selection to choose the most relevant features.Secondly,a data balancing technique is employed under a stratified cross-validation framework to address the issue of class imbalance.Through the combined application of these two methods,the aim is to enhance model performance and interpretability.The results demonstrate that the Tomek Link under sampling technique under the stratified cross-validation framework significantly improves the performance of various models.Additionally,successfully applying this method to the TabNet model also yields favorable results.Therefore,this comprehensive ap-proach holds practical value for predicting telecom customer churn,with the potential to improve churn detection rates and enhance business decision-making.