Telecommunications Churn Strategy

Research
Classification Modeling
Business Analytics
Poster project using classification models to identify churn risk and translate model results into retention strategy.
Published

December 31, 2024

Telecommunications Churn Strategy

Poster project using classification models to identify churn risk and translate model results into retention strategy.

Executive Summary

This project analyzed customer service records from a fictitious telecommunications company to identify which customer profiles were most strongly associated with churn. Using account, billing, and service subscription variables, I compared classification models and used the stronger model to estimate churn risk, identify key predictors, and translate model output into a targeted retention strategy.

The project connects predictive modeling with business decision-making. Rather than stopping at model performance, the analysis considered how churn probabilities, customer value, and outreach costs could be used together to prioritize retention resources.

Project Highlights

  • Compared logistic regression and decision tree models using cross-validation, with logistic regression achieving stronger predictive performance.
  • Identified customer characteristics associated with higher churn risk, including month-to-month contracts, electronic check payments, no tech support plan, and higher total charges.
  • Developed a retention strategy framework by combining predicted churn probability, customer billing value, outreach cost, and expected reduction in churn risk.

Methods & Approach

The analysis used customer account, billing, and service subscription variables to predict churn. Because the outcome was moderately imbalanced, I evaluated model performance using ROC-AUC and compared logistic regression against a decision tree approach. Logistic regression performed slightly better and was used to interpret the strongest predictors of churn.

After estimating churn probabilities, I translated the classification results into a practical retention strategy. Customers were segmented by both predicted churn risk and billing value to identify high-risk, high-value accounts. I then considered a simple expected-value scenario in which outreach costs $50 per customer and reduces churn probability by 20%, allowing the model results to inform which customers would be most worthwhile to target.

Deliverables

Research Poster
A poster presentation summarizing the churn modeling process, key predictors of customer attrition, risk classification strategy, and business implications for targeted retention outreach.

View Poster

GitHub Repository
Source code for the churn analysis, including model fitting, validation, visualization, and retention strategy calculations.

View Repository

Skills

Classification Modeling Logistic Regression Business Analytics Model Interpretation Data Visualization