【摘要】 构建Stacking融合模型，对电信高端客户的流失情况进行分析与预测。首先，根据用户生命周期理论将原始数据集划分为良好、行动和流失3个阶段，并利用充值金额筛选出高端客户；其次，对数据集进行主成分分析(PCA)、变量相关性分析等预处理和特征工程；然后，将处理后的数据集分别导入6个单一分类模型中，并结合5个评估指标对各模型的泛化性能进行比较。结果表明，单一分类模型的预测效果不甚理想，遂构建双层Stacking融合模型。通过评估指标判定，融合后模型的各指标数值均超过0.9,性能表现最佳。研究为Stacking融合模型分析客户流失问题提供了实证依据。更多还原

【Abstract】 A Stacking ensemble model was constructed to analyze and predict churn among high-end telecom customers. First, based on the Customer Lifecycle Theory, the original dataset was divided into three stages—retained, active, and churned—and high-end customers were identified using top-up amount thresholds. Second, data preprocessing and feature engineering were performed, including Principal Component Analysis(PCA) and variable correlation analysis. Then, the processed dataset was input into six individual base classification models, and their generalization performance was compared using five evaluation metrics. The results show that the predictive performance of the individual models was suboptimal. Therefore, a two-layer Stacking ensemble model was developed. According to the evaluation metrics, all performance indicators of the integrated model exceeded 0.9, demonstrating superior predictive capability. This study provides empirical evidence for applying Stacking ensemble models to customer churn analysis.更多还原