CUSTOMER AND PRICE ANALYSIS FOR CUSTOMER RETENTION USING CHURN MODELLING – Survey Paper

CUSTOMER AND PRICE ANALYSIS FOR CUSTOMER RETENTION USING CHURN MODELLING - Survey Paper

Mayuri Vengurlekar,Akash Varade,Athurv Bhogate,Aditya Swami,Swapnil Atpadkar

Dept.of Computer Engineering

K J College Of Engineering and Management Pune

Abstract—Customer churn prediction remains a critical challenge across diverse sectors, as retaining existing customers is more cost-effective than acquiring new ones. This survey provides a comprehensive review of machine learning (ML) and artificial intelligence (AI) techniques for churn prediction, drawing insights from industries including streaming services, telecommunications, banking, and retail. In streaming services, hybrid models have emerged as powerful tools. Techniques that combine deep learning architectures like Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) with traditional machine learning, such as Light Gradient Boosting (LightGBM), have proven effective. These models utilize sequential and static data to improve accuracy. A notable model achieved 95.60% AUC and an F1 score of 90.09% by integrating LSTM-GRU networks with LightGBM, using feature selection methods like Chi-squared testing and Sequential Feature Selection (SFS) to enhance predictive performance. In telecommunications, decision trees and ensemble models like Random Forests and XGBoost have been extensively used. These approaches excel in interpretability and predictive power. Studies demonstrate that Random Forest models consistently outperform Decision Trees, achieving higher AUC scores and accuracy metrics. The analysis incorporates tools such as confusion matrices and ROC curves to confirm these findings, highlighting Random Forests as reliable models for churn prediction. In the banking sector, models such as Logistic Regression, Support Vector Machines (SVM), and Artificial Neural Networks (ANN) have effectively predicted churn by analyzing customer transactions and behavior. Random Forest classifiers have reached accuracies up to 95.16%, especially when combined with data balancing methods like SMOTE. Moreover, Stacking models, which combine multiple classifiers, offer even better performance, surpassing CUSTOMER AND PRICE ANALYSIS FOR CUSTOMER RETENTION USING CHURN MODELLING1

traditional models like Random Forest and XGBoost.For the retail industry, predicting Customer and Price Analysis for Customer Retention using Churn Modelling A PREPRINT churn is crucial to implementing effective retention strategies. By using RFM (Recency, Frequency, Monetary) analysis and clustering techniques, businesses can anticipate shifts in customer behavior, allowing preemptive action. Integrating k-means clustering with predictive models improves the ability to identify at-risk customers earlier. The survey concludes that while traditional models like Decision Trees remain relevant, ensemble methods and hybrid models integrating deep learning are increasingly essential. Future research should focus on real-time processing and enhanced model interpretability to maximize business benefits across various sectors.

Index Terms—Customer churn, Machine learning, Churn prediction, Classification models, Neural Networks, Retail, Telecommunications, Banking, Streaming Services.