Customer Churn Prediction

An end-to-end machine learning project to predict customer churn and optimize recall using business-driven evaluation and threshold tuning.

---

Problem Statement

Customer churn is a major challenge for subscription-based businesses.
The objective of this project is to predict whether a customer is likely to churn and prioritize recall to minimize customer loss.

---

Key Insights from EDA

• Customers with short tenure churn significantly more
• Higher monthly charges are associated with higher churn
• Month-to-month contracts have the highest churn rate
• Lack of tech support / online security increases churn
• Dataset is imbalanced (~26% churn) → recall is more important than accuracy

---

Feature Engineering

• Converted TotalCharges to numeric and handled missing values
• Dropped non-informative customerID
• One-hot encoded categorical features
• Removed multicollinearity caused by redundant service features
• Scaled numerical features using StandardScaler
• Used stratified train-test split to preserve churn distribution

---

Modeling Approach

Logistic Regression (Baseline)

• Used as an interpretable baseline model
• Performed threshold tuning to optimize recall

Threshold Tuning

Instead of using the default 0.5 threshold, multiple thresholds were tested:

Threshold	Churn Recall
0.5	~55%
0.4 (Final)	~67%
0.3	~74% (too many false positives)

Final threshold chosen: 0.4

---

Model Comparison

A Random Forest model was trained to capture non-linear interactions.

Result:

• Random Forest achieved higher accuracy
• Logistic Regression achieved better recall for churn
• Logistic Regression chosen as final model due to business alignment and interpretability

---

Final Conclusion

This project demonstrates that model success depends on business objectives, not just accuracy.
By tuning the decision threshold and prioritizing recall, a simpler Logistic Regression model outperformed a more complex model for churn prediction.

---

Tech Stack

• Python
• Pandas, NumPy
• Matplotlib, Seaborn
• Scikit-learn
• Jupyter Notebook

---

Future Improvements

• Cost-sensitive learning
• Gradient Boosting (XGBoost / LightGBM)
• Deployment with Streamlit or FastAPI

---

Project Structure

Customer_Churn_Prediction/ ├── Data/ ├── Notebooks/ ├── README.md ├── requirements.txt └── .gitignore