Customer churn is a critical problem in the telecom industry, directly impacting revenue and customer lifetime value.

This project builds an end-to-end churn analysis and prediction system that:

• Identifies key churn drivers
• Segments high-risk customers
• Predicts churn using machine learning
• Provides actionable business recommendations

The goal is to analyze customer behavior and answer:

👉 Which customers are likely to churn, and what should the business do to retain them?

• Python
• Pandas & NumPy – Data analysis and manipulation
• Matplotlib & Seaborn – Data visualization
• Scikit-learn – Machine Learning (Logistic Regression)
• Jupyter Notebook – Analysis workflow

• Source: Kaggle – Telco Customer Churn Dataset
• Records: ~7,000 customers
• Features: Demographics, services, billing, and churn

Churn Risk Score	Churn by Tenure
Churn by Contract	Correlation Heatmap

1. Data Cleaning & Preprocessing
2. Exploratory Data Analysis (EDA)
3. Feature Engineering (customer segmentation)
4. Churn Risk Scoring
5. Machine Learning Model (Logistic Regression)
6. Business Insights & Recommendations

• Customers with low tenure (< 12 months) have the highest churn risk
• Month-to-month contracts significantly increase churn
• Higher monthly charges are associated with increased churn
• Customers using electronic check payments show higher churn
• High-risk customers are ~20x more likely to churn compared to low-risk

A Logistic Regression model was trained to predict churn.

• Accuracy: ~78.7%
• Precision (Churn): 0.62
• Recall (Churn): 0.51
• F1 Score: 0.56

=> The model is effective overall but has moderate recall, indicating room for improvement in detecting all churn cases.

A rule-based churn scoring system was created using:

• Low tenure (< 12 months)
• High monthly charges (> 70)
• Month-to-month contract

=> Customers with a score of 3 show ~69% churn, compared to ~3% for low-risk customers

• Fiber optic internet service
• Electronic check payment method
• Higher monthly charges

• Longer tenure
• One-year and two-year contracts

• Focus on customers within the first 12 months
• Improve onboarding and engagement

• Incentivize customers to switch to long-term plans

• Use churn scoring to prioritize retention efforts

• Improve value perception for high-cost plans

• Promote automatic payment methods

telecom-customer-churn-eda/
│
├── telecom_churn_analysis.ipynb
|
├── data/
│ └── Telco-Customer-Churn.csv
│
├── notebooks/
│ └── telecom_churn_analysis.ipynb
│
├── src/
│ ├── preprocessing.py
│ ├── feature_engineering.py
│ └── model.py
│
├── outputs/
│ ├── plots/
| | ├── churn_by_tenure.png
│ | └── churn_score.png
│ | └── churn_by_charges.png
│ | └── churn_by_contract.png
│ | └── heatmap.png
| |
│ └── model_results/
|   ├── feature_importance.csv
│   └── model_metrics.csv
│
├── images/
|   ├── churn_by_tenure.png
│   └── churn_score.png
│   └── churn_by_contract.png
│   └── heatmap.png
│
├── requirements.txt
└── README.md

1. Clone the repository

git clone https://github.com/Darshita-dp/telecom-customer-churn-eda.git

2. Navigate to the project: 
   cd telecom-customer-churn-eda
   pip install -r requirements.txt

3. Open Notebook:
   Jupyter Notebook

Improve model performance using Random Forest / XGBoost Increase recall for churn prediction Build interactive dashboard (Power BI / Streamlit) Deploy model as a web application

This project demonstrates how data analysis and machine learning can be combined to:

Identify churn drivers Predict high-risk customers Support data-driven retention strategies

=> The system can help businesses proactively reduce churn and improve long-term customer value.