A Study on Customer Churn of Commercial Banks Based on Learning

Oct 16, 2026

Uploaded by Fahsai

Paper investigates the problem of predicting customer churn for commercial banks using historical transaction data of customers. It addresses the high and prohibitive costs of acquiring labeled churn samples. A learning method is proposed from label proportions, where training data provides only group-level class proportions rather than individual labels. The group proportion is treated as estimable via customer risk experience, enabling cost reduction. The study applies Logistic Regression and Support Vector Machine for supervised learning from label proportions, and adds a genetic algorithm to solve an integer programming formulation, then compares results.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages7

Reading Time18 min

AccessFree to download

Introduction

Related Work

Customer churn prediction

Experimental Results

FAQs

What problem does the paper address?

It studies predicting customer churn for commercial banks using customers’ historical transaction data.

Why is label acquisition considered challenging in this study?

Obtaining labeled churn samples is described as costly and difficult, making conventional supervised approaches hard to apply.

What learning setting does the paper use?

It uses learning from label proportions, where training data contains only group-level label proportions while individual labels are treated as unknown.

What methods are used to solve the proposed learning tasks?

The paper uses Logistic Regression and Support Vector Machine for learning from label proportions, and also employs a genetic algorithm to solve an integer programming formulation for learning from label proportions.

Oct 16, 2026

Uploaded by Fahsai