MACHINE LEARNING BASED TELECOM-CUSTOMER CHURN PREDICTION AND OPTIMIZING CUSTOMER RETENTION

Abstract

Customer churn, or attrition, represents the percentage of customers who cease using
a company's services within a specified period. Calculated by dividing the number of
lost customers by the number retained at the period's outset, churn rate serves as a
critical metric for assessing a business's long-term viability. This research delves into
the application of machine learning and deep learning techniques to predict telecom
customer churn. Established methodologies like Random Forest Classifiers and
Support Vector Machines (SVMs) are juxtaposed against newer architectures such as
XGBoost and Deep Neural Networks (DNNs) to discern their efficacy in classifying
customer churn propensity. Furthermore, the models undergo rigorous evaluation
through grid search optimization. Results indicate that the Random Forest model
outperforms others for this specific use case, boasting a prediction accuracy of 90.96%
on testing data prior to grid search. This study not only sheds light on the predictive
capabilities of various algorithms but also underscores the significance of proactive
churn management in sustaining business growth and profitability.