Machine Learning Methods for Bank Term Deposit Subscription Prediction

Abstract

To address the problem of customer purchase behavior prediction in bank term-deposit services,  this paper develops a machine-learning-based prediction framework using the bank marketing dataset. First, the raw data are processed through missing-value inspection, categorical variable encoding, relevant feature selection, imbalance handling, and data standardization. Then, three models, namely Random Forest, Logistic Regression, and LightGBM, are constructed to analyze customer attributes, marketing-related attributes, and economic background attributes. Experimental results show that, among the three models, LightGBM achieves the best overall performance, with an accuracy of 91.95% and a test F1-score of 60.22%, outperforming both Random Forest and Logistic Regression. Further analysis indicates that features such as call duration and campaign frequency have strong influence on customer subscription decisions. The results demonstrate that machine learning methods can effectively improve the accuracy of identifying potential term-deposit customers, and provide useful data support for precision marketing and resource optimization in banking services.