ChurnRadar is a data-driven project focused on forecasting customer churn for a telecom provider. The goal is to uncover patterns in customer behavior and build a robust model capable of identifying customers at risk of leaving the service.
This project began as a hands-on application of data analysis and machine learning to a real-world problem. Along the way, it revealed not only statistical trends but also the subtle cues that often precede a customer’s departure.
- Source: Telco Customer Churn – Kaggle
- Contents: Customer demographics, account information, service usage, and churn status
- Size: 7,043 rows × 21 features
- Handled missing values and standardized data types
- Applied one-hot encoding to categorical variables
- Produced a clean dataset ready for modeling
- Explored churn rates by contract types, payment methods, service subscriptions, and more
- Used insightful visualizations such as boxplots and heatmaps to discover patterns
- Key observations:
- Month-to-month contracts show the highest churn
- Customers without online backup or tech support tend to churn more
- Even some customers streaming movies heavily still churned, indicating potential service issues
- Developed and compared Logistic Regression and Random Forest classifiers
- Split data into training and test sets
- Evaluated using precision, recall, and F1 score
- Logistic Regression performed best on this dataset
| Metric | Value |
|---|---|
| Precision | 66% |
| Recall | 56% |
| F1 Score | 60% |
| Overall Accuracy | 81% |
- Python
- Pandas, NumPy
- Seaborn, Matplotlib
- Scikit-learn
- Automatic payment methods (credit card, bank transfer) correlate with lower churn—possibly because these customers are more tech-savvy or trustful
- Tenure, monthly charges, and contract type strongly predict churn
- Lack of support services like online security and backup often signals customer dissatisfaction or unmet needs
ChurnRadar/
- ChurnRadar.ipynb
- README.md
- visuals/
- requirements.txt
- data/
Hi, I’m Saksham — currently pursuing my final year in Computer Science with a strong interest in real-world applications of AI and Machine Learning. This project was born out of a curiosity to understand why users leave services — not just from a modeling point of view, but from a behavioral lens. I'm especially drawn to data storytelling, clean visualizations, and building things that go beyond just accuracy scores.
If this project sparks ideas or you’d like to collaborate on something similar, I’d love to connect.