基于Stacking算法的银行定期存款产品购买行为研究

Purchase Behavior of Bank Time Deposit Products Based on Stacking Algorithm

摘要: 研究客户的购买行为以及客户的价值成为提高银行收益与优化营销策略的主要途径。本文通过研究客户购买某商业银行定期存款产品的相关数据，预测客户是否会购买该产品。使用LabelEncoding编码方法和SMOTE算法对数据进行处理。使用RFECV和GBDT算法进行特征选择，根据特征重要性获得影响客户是否购买的重要指标。使用决策树、SVM与GBDT算法以及Stacking算法对银行客户是否会购买定期存款产品进行研究，结果显示Stacking算法的预测效果比单一模型的预测效果更好。

Abstract: The research of customer’s purchase behavior and customer’s value has become the main way to improve bank’s income and optimize the marketing strategy. This paper studies the relevant data of customers’ purchase of fixed deposit products of a commercial bank to predict whether custom-ers will buy this product. SMOTE algorithm and Label Encoding were used to process the data. RFECV and GBDT algorithms are used for feature selection, and important indicators affecting whether customers buy are obtained according to the importance of features. Decision tree, SVM, GBDT algorithm and Stacking algorithm were used to study whether bank customers would buy time deposit products. The results showed that the prediction effect of Stacking algorithm was bet-ter than that of a single model.