📊 Customer Churn Prediction

An end-to-end Machine Learning project that predicts whether a telecom customer is likely to churn using customer demographics, account information, and service usage patterns.

🚀 Features

• Data Cleaning & Preprocessing
• Exploratory Data Analysis (EDA)
• Feature Engineering
• Machine Learning Pipeline using Scikit-Learn
• Logistic Regression
• Random Forest Classifier
• XGBoost Classifier
• Feature Importance Analysis
• Interactive Streamlit Web Application
• Model Serialization using Joblib

---

📂 Dataset

Dataset: Telco Customer Churn Dataset

The dataset contains customer information such as:

• Gender
• Senior Citizen Status
• Partner & Dependents
• Contract Type
• Internet Service
• Monthly Charges
• Total Charges
• Tenure
• Payment Method
• Churn Status

Target Variable:

Churn
0 -> Customer Stays
1 -> Customer Leaves

---

🔍 Exploratory Data Analysis

Key findings from EDA:

Contract Type

Contract	Churn Rate
Month-to-month	42.7%
One Year	11.3%
Two Year	2.8%

Month-to-month customers were significantly more likely to churn.

Tenure

Customers who churned generally had much lower tenure than retained customers.

Monthly Charges

Customers with higher monthly charges showed a higher tendency to churn.

---

🛠️ Data Preprocessing

Numerical Features

• SeniorCitizen
• tenure
• MonthlyCharges
• TotalCharges

Processing:

• Missing Value Imputation
• Standard Scaling

Categorical Features

• Gender
• Partner
• Dependents
• Internet Service
• Contract
• Payment Method
• Other service-related attributes

Processing:

• Missing Value Imputation
• One-Hot Encoding

Implemented using:

ColumnTransformer
Pipeline

---

🤖 Models Trained

Logistic Regression

• Baseline Model
• Fast and interpretable

Random Forest Classifier

• Ensemble Learning
• Bagging-based approach

XGBoost Classifier

• Gradient Boosting Framework
• Handles complex feature interactions

---

📈 Model Performance

Model	Accuracy	Precision	Recall	F1 Score
Logistic Regression	79.53%	63.75%	53.30%	58.06%
Random Forest	77.68%	60.71%	45.45%	51.99%
XGBoost	79.62%	64.65%	51.52%	57.34%

Best Performing Model

Logistic Regression achieved the highest F1 Score and Recall, making it the preferred model for deployment.

---

📊 Feature Importance (XGBoost)

Top churn-driving features:

1. Contract_Month-to-month
2. InternetService_Fiber optic
3. TechSupport_No
4. Contract_Two year
5. OnlineSecurity_No
6. Tenure
7. MonthlyCharges

These insights aligned closely with findings from the exploratory data analysis.

---

🌐 Streamlit Application

The project includes a Streamlit-based web interface where users can:

• Enter customer details
• Predict churn probability
• Identify customers at risk of leaving

Run locally:

streamlit run app.py

---

📁 Project Structure

customer-churn-prediction/
│
├── app.py
├── requirements.txt
├── README.md
│
├── data/
│   └── Dataset.csv
│
├── models/
│   └── churn_pipeline.pkl
│
├── notebooks/
│   └── churn_analysis.ipynb
│
├── src/
│   └── train.py
│
└── screenshots/

---

⚙️ Installation

Clone the repository:

git clone https://github.com/your-username/customer-churn-prediction.git

Move into the project directory:

cd customer-churn-prediction

Install dependencies:

pip install -r requirements.txt

---

🧪 Technologies Used

• Python
• Pandas
• NumPy
• Scikit-Learn
• XGBoost
• Streamlit
• Joblib
• Matplotlib

---

🎯 Future Improvements

• Hyperparameter Tuning
• ROC-AUC Analysis
• SHAP Explainability
• Cross Validation
• Docker Deployment
• Cloud Deployment (Render / Streamlit Cloud)

---

👨‍💻 Author

Arpit Shirbhate

If you found this project useful, consider giving it a ⭐ on GitHub.