Predicting Churn to Drive Retention Strategy for Telecom Clients
This end-to-end machine learning project focuses on predicting customer churn using a telecom dataset. With over 7,000 customer records, the project dives deep into patterns associated with churn behavior, applying a variety of classification algorithms to uncover actionable business insights.
Goal: To build and evaluate multiple ML models that can identify customers likely to churn and help businesses design effective retention strategies.
-
Comprehensive EDA & Preprocessing
→ Handled missing values, indirect missingness, outliers, label encoding, and feature scaling. -
Business Insights via Visualizations
→ Created advanced visualizations (donut charts, histograms, density plots, heatmaps) revealing key churn patterns like contract type, internet service, payment methods, tenure, tech support, and more. -
Modeling Techniques
→ Implemented and evaluated a suite of models:- K-Nearest Neighbors (KNN)
- Support Vector Machine (SVM)
- Random Forest
- Logistic Regression
- Decision Tree
- XGBoost (as AdaBoost alternative)
-
Evaluation Metrics
→ Accuracy, Confusion Matrix, ROC Curve, and Classification Reports were used to select the best-performing model.
- Language: R
- Libraries:
ggplot2,plotly,caret,randomForest,xgboost,e1071,glmnet,rpart,pROC,ROSE - Visualization: Plotly, Ggplot2, Ggcorrplot
- Data Handling: Dplyr, Tidyr, Data.table
- Customers on month-to-month contracts are 5x more likely to churn.
- Those with fiber-optic internet and electronic checks show high churn risk.
- Lack of tech support, online security, and dependents strongly correlates with churn.
- New customers (low tenure) are significantly more likely to churn.
This project empowers telecom providers to:
- Preemptively target at-risk customers with retention offers.
- Revise billing models or offer benefits for longer contracts.
- Improve customer service by investing in tech support and online services.
📂 Customer-Churn-Prediction/
│
├── Group 10 Project.R # End-to-end R code
├── Project Outputs.docx # EDA results and model evaluations
├── Customer Churn Prediction (REPORT).docx # Final project report
├── Customer_Churn_Prediction.pptx # Presentation slide deck
├── Customer_Telecom_Data.csv # Input dataset
└── README.md # This fileSurya Vamsi Patiballa
M.S. in Data Science – George Washington University