Document Type
Thesis
Publication Date
4-23-2026
Abstract
This capstone project addresses a core business problem in the telecommunications industry: how to detect, measure, and prevent customer churn before it occurs, using a KPI framework validated by statistical analysis and machine learning. Applied to 7,043 customer records across 37 variables from a California-based telecom operator, the analysis integrates descriptive KPI computation, Pearson correlation analysis, chi-square tests of independence, and logistic regression churn prediction. The research confirms that customer tenure (r = −0.352, p < 0.001) and contract type (Cramér's V = 0.410, p < 0.001) are the dominant predictors of churn, and that a logistic regression model achieves AUC-ROC = 0.8307 — sufficient for proactive deployment. A 5-percentage-point churn reduction preserves $273,670 in annual revenue and $1.55M in customer lifetime value. The project delivers an integrated analytical framework, a predictive retention model, a 16-chart visualization suite, and a 12-page Power BI executive dashboard.
Recommended Citation
Malik, Moeez Islam '27 and Eremionkhale, Amy Ehinomen, "Predicting Telecom Customer Churn: A Statistically Validated KPI Framework with Machine Learning and Revenue Impact Modeling" (2026). Student Research. 634, Scholarly and Creative Work from DePauw University.
https://scholarship.depauw.edu/studentresearchother/634
Comments
Completed as part of BUSA485 Independent Capstone Project with Professor Amy Eremionkhale. Co-Authored/Revised by Professor Amy Eremionkhale.