Customer Churn Prediction

A comprehensive machine learning pipeline for predicting customer churn in the telecommunications industry.

Overview

This repo contains an end-to-end solution for predicting customer churn, helping businesses identify customers at risk of leaving their service. The pipeline implements best practices in data preprocessing, feature engineering, model training, and evaluation.

Features

• Robust Data Preprocessing: Handles categorical variables, missing values, and data transformations for 7,043 customer records
• Advanced Feature Engineering: Creates 9 powerful derived features that capture customer behavior patterns
• Cross-Validated Model Training: Ensures model reliability and stability
• Comprehensive Evaluation: Multiple metrics including accuracy, precision, recall, F1-score, and AUC-ROC
• Insightful Visualizations: Various plots to understand churn drivers and model decisions
• Production-Ready Prediction: Returns probability, risk level, and binary prediction

Pipeline Components

1. Data Preprocessing (data_preprocessing.py)

   • Standardizes data formats
   • Converts categorical variables to numeric
   • One-hot encodes multi-class categorical features

2. Feature Engineering (feature_engineering.py)

   • Creates basic and advanced derived features:
     • AvgMonthlyCharge: Average charge per month
     • HighValueFlag: Identifies high-value customers
     • TenureToChargeRatio: Relationship between tenure and charges
     • ServiceDiversity: Variety of services used
     • TenureByContract: Interaction between tenure and contract type
     • TotalServices: Count of services subscribed
     • ServiceDensity: Service concentration
     • PotentialLTV: Estimated lifetime value
     • ChurnRiskScore: Aggregated risk score

3. Model Training (model_training.py)

   • Implements machine learning model with cross-validation
   • Current performance metrics:(still tuning)
     • Accuracy: 76%
     • Precision: 53%
     • Recall: 73%
     • F1-Score: 61%
     • AUC-ROC: 84%

4. Visualization (visualization.py)

   • Creates multiple visualizations:
     • Contract vs Churn
     • Monthly Charges vs Churn
     • Tenure vs Churn
     • Internet Service vs Churn
     • Payment Method vs Churn
     • Correlation heatmap
     • SHAP visualizations for model explainability

Installation

# Clone the repository
git clone https://github.com/MessoJ/Customer-Churn-Prediction.git
cd Customer-Churn-Prediction

# Install dependencies
pip install -r requirements.txt

Usage

Running the Full Pipeline

# Run the full pipeline
python data_preprocessing.py
python feature_engineering.py
python model_training.py
python model_validation.py
python model_evaluation.py
python visualization.py
python prediction.py

Model Insights

The current model prioritizes recall (73%) over precision (53%), making it well-suited for identifying as many potential churners as possible, even at the cost of some false positives. This approach is business-oriented, as the cost of missing a potential churner typically exceeds the cost of incorrectly flagging a loyal customer.

Key findings from the analysis:

• Contract type is a strong predictor of churn (month-to-month contracts have higher churn)
• Fiber optic internet service users show higher churn rates
• Electronic check payment method is associated with higher churn
• Short tenure strongly predicts churn likelihood

Project Structure

Customer-Churn-Prediction/
├── data/
│   └── telco_churn.csv
├── data_preprocessing.py
├── feature_engineering.py
├── model_evaluation.py
├── model_training.py
├── model_validation.py
├── prediction.py
├── visualization.py
├── utils.py
├── requirements.txt
└── README.md

What I'm working on to improve, You can also contribute

• I'm mplementing hyperparameter tuning
• Exploring ensemble models (Random Forest, XGBoost)
• Adding more interaction features
• Developing customer segmentation before classification

License

MIT

Acknowledgments

This project (is) and was developed using telecommunication customer data to help businesses improve customer retention strategies.