Customer churn is a major challenge for telecom companies because losing existing customers can negatively impact recurring revenue and increase customer acquisition requirements.
This project analyzes telecom customer data to identify customer churn patterns, high-risk segments, and factors associated with customer attrition.
The project combines Python-based exploratory and statistical analysis with Logistic Regression and an interactive Power BI dashboard to translate customer data into actionable business insights.
Business Objectives
The analysis aims to answer:
What is the overall customer churn rate? Which contract types have the highest churn? How does customer tenure relate to churn? Which internet service categories show higher churn? Which payment methods are associated with higher churn? How does monthly charging relate to churn? Which customer segments should be prioritized for retention?
Dataset: IBM Telco Customer Churn Dataset
The project uses a publicly available telecom customer churn dataset containing information on:
- Customer demographics
- Service subscriptions
- Contract information
- Billing details
- Payment methods
- Customer tenure
- Churn status
| Tool / Technology | Purpose |
|---|---|
| Python | Data analysis, exploratory analysis & predictive modelling |
| Pandas | Data cleaning, transformation & manipulation |
| NumPy | Numerical operations & data processing |
| Matplotlib | Data visualization |
| Seaborn | Statistical data visualization & pattern analysis |
| Scikit-learn | Logistic Regression, model training & evaluation |
| Power BI | Interactive dashboard development & business reporting |
| DAX | KPI calculations & analytical measures |
| GitHub | Project documentation & version control |
The telecom customer dataset was imported and prepared for analysis.
Key preprocessing steps included:
- Preparing categorical variables for modelling
- Creating dummy variables for categorical features
- Defining independent and dependent variables
- Splitting the dataset into training and testing sets
Customer churn was analyzed across multiple dimensions, including:
- Contract type
- Customer tenure
- Internet service
- Payment method
- Monthly charges
- Customer characteristics
- Additional subscribed services
The analysis was used to identify customer segments with relatively higher and lower churn rates.
A Logistic Regression model was developed to estimate the likelihood of customer churn.
The model was evaluated using:
- Accuracy
- Precision
- Recall
- Confusion Matrix
- ROC-AUC
| Metric | Result |
|---|---|
| Accuracy | 79.45% |
| Precision | 62.24% |
| Recall | 55.60% |
| ROC-AUC | 84.43% |
The model achieved an ROC-AUC of 84.43%, indicating good ability to distinguish between customers who churn and those who do not.
However, the recall of 55.60% indicates that there is scope to improve the identification of actual churners, particularly when minimizing missed churn-risk customers is a business priority.
| Predicted: No Churn | Predicted: Churn | |
|---|---|---|
| Actual: No Churn | 1,827 | 250 |
| Actual: Churn | 329 | 412 |
The model correctly identified 412 churned customers, while 329 actual churners were not identified by the model.
The Logistic Regression coefficients were examined to understand the direction of association between customer attributes and churn.
Some of the notable positive coefficients were:
| Variable | Coefficient |
|---|---|
| Month-to-month | 0.984 |
| Fiber optic | 0.547 |
| One year | 0.520 |
| Paperless Billing | 0.239 |
| Multiple Lines | 0.216 |
| Electronic check | 0.299 |
Notable negative coefficients included:
| Variable | Coefficient |
|---|---|
| Phone Service | -1.314 |
| Online Security | -0.546 |
| Tech Support | -0.439 |
| Online Backup | -0.293 |
| Device Protection | -0.263 |
| Tenure | -0.037 |
These coefficients indicate associations within the fitted model and should not be interpreted as causal effects.
An interactive Power BI dashboard was developed to provide a business-focused view of customer churn.
- Total Customers
- Churned Customers
- Overall Churn Rate
- Average Monthly Charges
- Average Customer Tenure
The dashboard analyzes churn by:
- Contract Type
- Internet Service
- Customer Tenure
- Monthly Charges
- Payment Method
Interactive slicers allow users to filter the analysis by key customer attributes.
Month-to-month customers show a 42.71% churn rate, compared with 11.27% for one-year contracts and 2.83% for two-year contracts.
This indicates that customers without long-term commitments represent an important retention segment.
Customers with 0–12 months of tenure have the highest churn rate at 47.44%.
Churn decreases as tenure increases, reaching 9.51% among customers with 49+ months of tenure.
This highlights the first year of the customer lifecycle as an important retention window.
Fiber optic customers show a 41.89% churn rate, compared with 18.96% for DSL and 7.40% for customers without internet service.
The higher churn observed among fiber optic customers warrants further investigation into pricing, service quality, customer experience and competitive alternatives.
Customers using electronic check have the highest churn rate at 45.29% among the payment methods analyzed.
This segment could be investigated further to understand whether payment experience or customer characteristics contribute to the higher observed churn.
Develop stronger onboarding, engagement and customer-support initiatives during the first 12 months, when churn is highest.
Develop targeted retention offers and incentives to encourage month-to-month customers to move toward one-year or two-year contracts.
Further investigate the customer experience, pricing, service quality and competitive environment associated with fiber optic customers.
Investigate whether electronic-check customers experience payment friction or represent a particular high-risk customer segment.
The model's 84.43% ROC-AUC demonstrates useful discriminatory ability, but the 55.60% recall indicates that additional modelling, feature engineering or threshold optimization could improve identification of customers at risk of churn.
Data Analysis | Data Cleaning | Exploratory Data Analysis | Statistical Modelling | Logistic Regression | Model Evaluation | Python | Pandas | NumPy | Scikit-learn | Power BI | DAX | Data Visualization | Customer Segmentation | Business Insights
Author
Swarnamayee Kar
Business Analyst | Data Analyst