Customer Churn Prediction System

Overview

Machine Learning system that predicts which customers are likely to leave, enabling proactive retention strategies and significant cost savings.

Business Impact

ROI: 621.3% - Every $1 invested returns $6.21 Annual Profit: $185,150 Churn Detection Rate: 79% - Catches 4 out of 5 customers who will leave Revenue Saved: $247,800 annually

Model Performance

Algorithm: Logistic Regression with Feature Selection ROC-AUC Score: 0.832 (Excellent performance) Recall: 79% (High churn detection rate) Precision: 49% (Balanced approach) Dataset Size: 7,043 real telecommunications customers

Technologies Used

Core Technologies:

• Python 3.8+
• pandas 2.0.3
• scikit-learn 1.3.0
• numpy 1.24.3

Visualization & Analysis:

• matplotlib 3.7.1
• seaborn 0.12.2
• Jupyter Notebook

Machine Learning Techniques:

• Classification algorithms
• Feature engineering and selection
• Cross-validation
• Model optimization

Dataset Features

Customer Demographics:

• Age, gender, partner status
• Dependents and family structure
• Geographic information

Service Usage Patterns:

• Tenure (months with company)
• Monthly charges
• Total charges paid
• Service utilization metrics

Contract Details:

• Contract type (monthly, yearly, two-year)
• Payment method
• Billing preferences
• Auto-payment status

Service Subscriptions:

• Internet service type
• Phone service options
• Streaming services
• Security and backup services

Key Business Insights

1. Financial Predictors: Total charges and monthly charges are the strongest churn indicators
2. Customer Lifecycle: New customers with low tenure represent the highest churn risk
3. Contract Structure: Month-to-month contracts show 3x higher churn rates than long-term contracts
4. Service Type Impact: Fiber optic customers demonstrate higher churn rates compared to DSL users

Project Structure

Customer_Churn_Prediction/ ├── Customer_Churn_Prediction_Project.ipynb # Main analysis notebook ├── README.md # Project documentation ├── requirements.txt # Python dependencies └── data/ └── dataset_info.md # Dataset documentation

Installation and Usage

Prerequisites:

• Python 3.8 or higher
• Jupyter Notebook or JupyterLab

Setup Instructions:

1. Clone the repository git clone https://github.com/yourusername/Customer_Churn_Prediction_Project.git cd Customer_Churn_Prediction_Project

2. Install required packages pip install -r requirements.txt

3. Launch Jupyter Notebook jupyter notebook

4. Open and run Customer_Churn_Prediction_Project.ipynb

Model Development Process

Data Preprocessing:

• Handled missing values in TotalCharges column
• Encoded categorical variables using LabelEncoder
• Applied StandardScaler for feature normalization
• Performed train-test split with stratification

Feature Engineering:

• Correlation analysis to identify key predictors
• Feature importance ranking using Random Forest
• Selection of top 8 most impactful features
• Removal of low-impact variables

Model Selection:

• Evaluated Logistic Regression, Random Forest, and SVM
• Optimized for business metrics rather than accuracy alone
• Applied class balancing to address churn imbalance
• Cross-validated performance metrics

Business Applications

Telecommunications Industry:

• Reduce customer acquisition costs
• Optimize retention campaign targeting
• Improve customer lifetime value

SaaS and Subscription Services:

• Identify at-risk subscribers
• Implement proactive retention strategies
• Reduce revenue churn

Financial Services:

• Prevent account closures
• Enhance customer relationship management
• Optimize retention investment allocation

Future Enhancements

Technical Improvements:

• Implementation of real-time prediction API
• Integration of deep learning models for improved accuracy
• Development of ensemble methods for robust predictions

Business Intelligence:

• A/B testing framework for retention campaigns
• Customer segmentation for personalized strategies
• Advanced feature engineering from customer behavior data

Operational Integration:

• Automated alert system for high-risk customers
• Dashboard for real-time churn monitoring
• Integration with CRM systems

Performance Metrics

Model Performance Comparison:

• Logistic Regression: Accuracy 73.0%, Precision 49%, Recall 79%, F1-Score 61%, ROC-AUC 0.832
• Random Forest: Accuracy 79.5%, Precision 64%, Recall 51%, F1-Score 57%, ROC-AUC 0.825
• SVM: Accuracy 79.3%, Precision 65%, Recall 48%, F1-Score 55%, ROC-AUC 0.791

Author

Tzipora Elbaz Machine Learning Engineer Specializing in predictive analytics and business intelligence

Contact

For questions or collaboration opportunities, please reach out through GitHub issues.

Note: This project demonstrates end-to-end machine learning pipeline development with focus on business impact and ROI optimization.