In recent years, the telecommunications industry has witnessed intensified competition, wherein the expense associated with acquiring new consumers exceeds that of sustaining existing ones. Consequently, predicting customer churn prior to its occurrence has become essential. This study proposes a sentiment-based customer churn prediction model in which the sentiment of customers is predicted using Random Forest. Subsequently, the derived sentiment predictions are combined with additional features to predict customer churn. The ensemble technique is applied to predict churn, consisting of K-nearest neighbors, Support Vector Machines, Random Forest as base learners, and Multiple Layer Perceptron as a meta learner. Moreover, mutual information is applied to select the pertinent features impacting customer churn, and the class imbalance is handled through the utilization of the class weighted technique. The results of the experiments reveal that the proposed model surpassed the state-of-the-art customer churn models as it achieved an accuracy of 98.86%, an AUC of 99.47 %, and an F1-score of 97.77%.


School of Sciences and Engineering


Computer Science & Engineering Department

Degree Name

MS in Computer Science

Graduation Date

Fall 2-8-2024

Submission Date


First Advisor

Ahmed Rafea

Second Advisor

Mona Farouk

Committee Member 1

Hossam Sharara

Committee Member 2

Abeer ElKorany


61 p.

Document Type

Master's Thesis

Institutional Review Board (IRB) Approval

Approval has been obtained for this item

Available for download on Saturday, August 17, 2024