🎵 Streaming Churn Prediction Model

📊 Project Description

This project develops a Machine Learning model to predict user churn in streaming services. The model uses advanced classification techniques to identify behavioral patterns that indicate subscription cancellation risk.

🎯 Objective

Identify the probability that customers will abandon the streaming service, enabling proactive retention strategies and optimizing marketing resources.

📈 Key Results

• AUC-ROC: 93.47% (Excellent predictive capability)
• Accuracy: 84.78% (High overall precision)
• Best Model: Random Forest
• Dataset: 125,000 users with 20 features

🏗️ Project Architecture

streaming-churn-prediction-model/
├── modelochurd.ipynb          # Main notebook with complete analysis
├── train.csv                  # Training dataset
├── test.csv                   # Test dataset
└── README.md                  # This file

Note: The trained models (best_rf_label.pkl, best_xgb_label.pkl, best_logistic_regression.pkl) are automatically generated when executing the complete notebook.

🔍 Data Analysis

Key Variables Identified:

1. weekly_hours - Weekly usage hours (most important)
2. customer_service_inquiries - Customer service inquiries
3. subscription_type - Subscription type
4. song_skip_rate - Song skip rate
5. num_subscription_pauses - Subscription pauses

Business Insights:

• Free users: 79.4% churn rate
• Premium/Family users: 34-35% churn rate
• Lower weekly usage = Higher churn risk
• More support inquiries = Higher abandonment probability

🤖 Implemented Models

Model	AUC-ROC	Accuracy	Precision	Recall	F1-Score
Random Forest	0.9347	84.78%	84.89%	85.60%	85.24%
Logistic Regression	0.8935	80.44%	80.97%	80.91%	80.94%
XGBoost	0.8732	77.24%	79.08%	75.68%	77.35%

🛠️ Technologies Used

• Python 3.x
• Pandas - Data manipulation
• Scikit-learn - Machine Learning
• XGBoost - Boosting algorithm
• Seaborn/Matplotlib - Visualizations
• NumPy - Numerical computation

📋 Dataset Features

Numerical Variables:

• age - User age
• weekly_hours - Weekly usage hours
• average_session_length - Average session duration
• song_skip_rate - Song skip rate
• weekly_songs_played - Songs played per week
• num_subscription_pauses - Number of subscription pauses
• customer_tenure_years - Customer tenure

Categorical Variables:

• subscription_type - Subscription type (Free, Premium, Family, Student)
• payment_plan - Payment plan (Monthly, Yearly)
• payment_method - Payment method
• location - User location
• customer_service_inquiries - Customer service inquiry frequency

🚀 Installation and Usage

Prerequisites:

pip install pandas numpy scikit-learn xgboost seaborn matplotlib

Execution:

1. Clone the repository
2. Open modelochurd.ipynb in Jupyter Notebook
3. Run all cells to reproduce the complete analysis
4. Los modelos entrenados se guardarán automáticamente como archivos .pkl

Using Trained Models:

import pickle

# Load model (se genera al ejecutar el notebook)
with open('best_rf_label.pkl', 'rb') as f:
    model = pickle.load(f)

# Make predictions
predictions = model.predict(X_new_data)
probabilities = model.predict_proba(X_new_data)

Importante: Los archivos .pkl de los modelos se crean automáticamente al ejecutar todas las celdas del notebook modelochurd.ipynb. Si no existen, ejecuta el notebook completo para generarlos.

📊 Evaluation Metrics

Primary Metric: AUC-ROC

• 0.5: Random performance
• 0.7-0.8: Good
• 0.8-0.9: Very good
• 0.9+: Excellent
• 1.0: Perfect

Complementary Metrics:

• Accuracy: Proportion of correct predictions
• Precision: Efficiency of positive predictions
• Recall: Ability to capture real cases
• F1-Score: Balance between precision and recall

💡 Business Applications

Retention Strategies:

1. User segmentation by churn risk
2. Personalized campaigns for high-risk users
3. Marketing resource optimization
4. User experience improvement

KPIs to Monitor:

• Churn rate by segment
• Retention campaign effectiveness
• Loyalty strategy ROI
• Customer satisfaction

🔬 Methodology

1. Exploratory Data Analysis (EDA)

   • Variable distribution
   • Correlations
   • Missing value analysis

2. Data Preparation

   • Categorical variable encoding
   • Numerical variable scaling
   • Feature engineering

3. Modeling

   • Multiple algorithm training
   • Hyperparameter optimization
   • Cross-validation

4. Evaluation

   • Model comparison
   • Feature importance analysis
   • Result interpretation

📈 Included Visualizations

• Target variable distribution
• Correlation matrices
• Feature importance
• Confusion matrices
• ROC and Precision-Recall curves
• Churn analysis by categories

📄 License

This project is under the MIT License. See the LICENSE file for more details.

👨‍💻 Author

Your Name

• LinkedIn: https://www.linkedin.com/in/jesus-beleno/
• Email: jesusbelenov@gmail.com

🙏 Acknowledgments

• Dataset provided by [https://www.kaggle.com/competitions/streaming-subscription-churn-model/team]
• Data Science community
• Open source tools used

---

⭐ If this project was helpful, please give it a star!