Customer Churn Prediction Using Deep Learning Techniques

Authors: Ghuge Shankar Ramesh, Shinde Aryan Anil, Vetal Om Ravindra, Sangale Saurabh Baban, R. N. Muneshwar

Country: India

Full-text Research PDF File:   View   |   Download

Abstract: The Churn Prediction System is designed to help companies identify customers who are likely to stop using their services in the near future. This project uses Machine Learning and Deep Learning techniques to analyze customer behavior and predict churn patterns effectively. The system allows an Admin to register, log in, and upload multiple company datasets. Once the datasets are uploaded, the system processes them using predictive models to determine which customers might leave or may not reorder in the next month. To achieve high prediction accuracy, multiple Machine Learning algorithms are implemented and evaluated, including Logistic Regression, Decision Tree, Support Vector Machine (SVM), and Random Forest. Among these, the Random Forest algorithm outperformed all other models with the highest accuracy of 94.6%, due to its ensemble learning approach and ability to handle complex data patterns. In comparison, Decision Tree achieved 88.3% accuracy, SVM achieved 86.9%, and Logistic Regression achieved 84.7% accuracy. This demonstrates that Random Forest provides more reliable and stable predictions for customer churn analysis. The results are displayed on an interactive dashboard with visual insights through bar charts, pie charts, and growth graphs, helping companies easily understand their churn trends. The system also provides a download option to export customer details (in Excel format) for those predicted to leave. Additionally, it generates feedback and improvement suggestions to help companies understand the reasons behind customer loss.

Keywords: Customer Churn Prediction, Machine Learning, Deep Learning, Data Analytics, Customer Retention Predictive Modeling.

Paper Id: 233034

Published On: 2026-04-03

Published In: Volume 14, Issue 2, March-April 2026

Share this