Transforming Marketing Analytics: A Machine Learning Framework for Churn Prediction Driven by Data

Abstract

Customer attrition is one of the main issues facing large corporations. Businesses in the rental sector 
especially look for ways to retain customers since they are their main source of revenue. The data-driven 
machine learning approach for telecom sector churn prediction is presented in this study. It tackles the problems 
of noisy, imbalanced, and high-dimensional data by using a thorough preprocessing pipeline that includes feature 
selection, outlier detection, SMOTE-based balancing, and noise removal. The preprocessed data, which includes 
3,333 customer records with variables ranging from call metrics to charge information, is divided into training 
and testing sets in order to develop prediction models utilising Decision Trees and Random Forest classifiers. 
The Decision Tree model outperformed other models like ANN and SVM when evaluated using metrics like 
accuracy, precision, recall, F1-score, and ROC analysis. It achieved 88% accuracy, 83% precision, 93% recall, 
and 88% F1-score. These results demonstrate how the method may provide valuable data for enhancing client 
retention and optimising marketing strategies in evolving telecom environments.