This project is focused on predicting customer churn using a dataset from a telecommunications company. The goal is to identify customers who are likely to stop using the service, and understand the business factors behind churn, enabling proactive retention strategies.
The main objective of this project is to:
- Analyze customer data to discover patterns and churn drivers
- Build machine learning models to predict churn
- Compare models to select the best performing one
- Translate insights into actionable business recommendations
- Source: Telco Customer Churn Dataset (from IBM Sample Datasets or Kaggle)
- Rows: 7043 customers
- Columns: 21 (features include customer demographics, services, payment info, etc.)
- Target:
Churn(Yes/No)
We explored the data using Seaborn and Matplotlib. Key visualizations include:
- ✅ Pie chart showing overall churn rate
- ✅ Histograms for tenure and monthly charges by churn
- ✅ Countplots for contract type and payment method vs churn
- ✅ Correlation heatmaps for numeric features
These helped us form business insights such as:
- Customers with short tenure churn more
- Month-to-month contracts are associated with higher churn
- Electronic check users churn more than credit card users
- Higher charges lead to higher churn risk
We trained and evaluated the following models:
| Model | Accuracy |
|---|---|
| Logistic Regression | ~80% |
| Random Forest | ~82% ✅ Best |
| Support Vector Machine (SVM) | ~78% |
| K-Nearest Neighbors | ~75% |
| Naive Bayes | ~72% |
- We used Label Encoding, Feature Scaling, and Train-Test Split.
- Models were evaluated using Accuracy, Confusion Matrix, and Classification Report.
- Best-performing model: Random Forest Classifier
All visualizations used for EDA and model comparison are saved in the visuals/ folder. These include:
- Churn Distribution Pie Chart
- Tenure Distribution by Churn
- Monthly Charges KDE by Churn
- Contract Type vs Churn
- Model Accuracy Bar Plot
This project doesn't just build a predictive model — it translates data into business value:
- ~27% of customers churn — a serious revenue leakage
- Customers with month-to-month contracts churn at 42%, vs just 11% on long-term contracts
- Short-tenure users (<12 months) are most likely to leave — onboarding needs improvement
- Electronic check as a payment method is a major churn signal
- Our model can help businesses predict churners in advance and run targeted retention campaigns
| Strategy | Explanation |
|---|---|
| 🎯 Target short-tenure users | Offer onboarding perks or welcome incentives |
| 📄 Encourage long-term contracts | Provide discounts to switch from month-to-month |
| 💳 Avoid churn-prone payment methods | Encourage auto-pay or credit card users |
| 📢 Use model predictions | Export churn scores weekly to customer support & marketing |
- Python
- Pandas, NumPy (data cleaning & processing)
- Seaborn, Matplotlib (visualizations)
- Scikit-learn (modeling)
- Jupyter Notebook
- Use SMOTE for class imbalance
- Deploy the model with a Streamlit app for business users
- Incorporate customer sentiment or feedback as additional features
- Track model performance over time with real churn data
I'm a third-year Computer Applications student with a passion for data science. This project reflects my ability to combine technical machine learning skills with real-world business thinking. Looking for roles in Data Analysis or Data Science.
📫 Reach me on LinkedIn
📂 Portfolio: My Portfolio ->
If you found this project useful, please give it a ⭐ on GitHub. It helps me grow!