基于SMOTE增强与多模型择优的银行客户忠诚度预测研究

Research on the Prediction of Bank Customer Loyalty Based on SMOTE Enhancement and Multi-Model Optimization

摘要: 本研究针对银行客户流失预测问题，通过系统性数据处理、可视化分析与特征工程，构建了多种机器学习模型(包括逻辑回归、随机森林、AdaBoost和支持向量机)，并基于ROC曲线、F1分数等指标评估模型性能。核心发现表明，随机森林模型在应对数据不平衡和捕捉复杂特征关系方面表现最优(测试集F1分数达0.8546)，显著优于其他模型；方法贡献在于提出了一套结合可视化探索与特征优化的建模框架，强调了数据质量与衍生特征对预测性能的关键作用；研究局限包括数据来源单一性及模型对特定业务场景的泛化能力有待进一步验证。本研究为银行客户忠诚度管理提供了数据驱动的决策支持。

Abstract: This study addresses the problem of bank customer churn prediction. By systematic data processing, visualization analysis, and feature engineering, a variety of machine learning (including logistic regression, random forest, AdaBoost, and support vector machine) are constructed and evaluated based on ROC curves, F1 scores, and other metrics. Core findings show that the random forest model performs the best in dealing with data imbalance and capturing complex feature relationships (achieving a F1 score of 0.8546 the test set), significantly outperforming the other models. The methodological contribution lies in proposing a modeling framework that combines visualization exploration and feature optimization, emphasizing the critical roles of quality and derived features in prediction performance. Research limitations include the singularity of data sources and the need for further validation of the model’s generalization capability in specific business scenarios. This study provides data-driven decision support for bank customer loyalty management.