Project Series
Predicting Churn in Telecom Services
-
Date: Apr 2023
-
Tools:
-
Power BI - Visualization
-
RapidMiner - Machine Learning Prediction
-
-
Data Volume: 7,043
-
Data Source: kaggle (Telco Customer Churn - IBM Sample Data Sets)
-
Project Outcome: Interactive Dashboard, Machine Learning model comparison
Project Summary
This project simulates a customer churn scenario to help businesses identify customers at risk of churning and the underlying factors contributing to their decision. The analysis is based on IBM's Telco Customer Churn dataset, sourced from Kaggle.
The study consists of two analytical components:
1. Exploratory Visualization (Power BI):
Interactive dashboards provide insights into customer profiles, overall churn patterns, and the relationships between churn rate and various factors.
2. Predictive Modeling (RapidMiner):
Two machine learning algorithms—Logistic Regression and Decision Tree—were employed to predict customer churn and uncover significant risk variables. Model performance was compared to determine the more effective approach.
Insights from Exploratory Visualization (Power BI):
-
Churn rate is associated with tenure, contract type, service type, and monthly charges.
-
By analyzing the interaction between tenure and monthly charges, the highest and lowest churn risk segments were identified:
🔺 High monthly charge + Low tenure → Highest churn rate
🔻 Low monthly charge + High tenure → Lowest churn rate
Model Performance:
-
Logistic Regression achieved an accuracy of 78.8%, outperforming the Decision Tree model at 75.93%.
-
Contract type, tenure, and internet service emerged as significant predictors in both models.
-
Monthly charges were significant only in the Decision Tree model.
Visualizations
Demo - Interactive Dashboard
Dashboards: Customer Profile and Churn Analysis
Predictive Modeling
Churn Prediction with Decision Tree
Churn Prediction with Logistic Regression
Conclusion
Churn is most prevalent among short-tenure customers with high monthly charges, and is also elevated among those with limited contract commitments. Logistic Regression achieved 78.8% accuracy, identifying tenure, contract type, and internet service as key churn predictors. Businesses can tailor retention strategies by targeting at-risk customer segments and optimizing service offerings based on these insights.