🎯 Telecom Customer Churn Prediction

Machine learning solution to predict customer churn and enable proactive retention strategies

---

🚀 Business Problem

Telecom operators face significant challenges with customer retention, as customer acquisition costs are typically 5-25x higher than retention costs. This project develops a machine learning solution to identify customers at risk of churning, enabling Interconnect to offer targeted promotional codes and special plans before customers leave.

💡 Why This Matters

• Proactive retention strategies through early churn identification
• Optimize marketing spend by targeting high-risk customers
• Reduce revenue loss from customer defection
• Increase customer lifetime value through data-driven interventions

---

📊 Dataset Overview

Multi-source dataset from Interconnect telecom operator including 7,043 customers:

Data Source	Key Features
📋 Contract Information	Begin/end dates, contract types, billing preferences, payment methods
👤 Personal Data	Demographics, senior citizen status, family information
🌐 Internet Services	DSL/Fiber connections, security features, streaming services
📞 Phone Services	Multiple line options and usage patterns

🎯 Target Variable

• 26.5% churn rate (1,869 churned customers out of 7,043)
• Business-critical imbalance requiring specialized handling

---

🔬 Methodology

🧹 Data Preprocessing & Engineering

• Multi-source integration: Merged 4 datasets using customer ID joins
• Date processing: Converted contract dates to identify active vs churned customers
• Missing value treatment: Filled gaps with business logic (new customers = 0 total charges)
• Feature engineering: Created duration_days from contract length for enhanced variability
• Categorical encoding: Mapped text values to numeric (Yes/No → 1/0, contract types → 0/1/2)

📈 Exploratory Data Analysis

Key business insights discovered:

Finding	Churn Impact
📝 Electronic check payments	Higher churn rates - potential payment friction
📅 Month-to-month contracts	Significantly higher churn vs annual contracts
💰 Monthly charges	Churned customers average ~$80 vs $65 for retained
🌐 Fiber optic service	Highest churn ratio - service quality issues
⏱️ Contract duration	Longer tenure strongly correlates with retention

🤖 Model Development

Implemented comprehensive ML pipeline:

# Models Tested
├── Dummy Classifier (Baseline)
├── Logistic Regression 
├── Random Forest
├── Gradient Boosting ⭐ (Best Performer)
└── AdaBoost

Advanced Techniques:

• SMOTE oversampling for class imbalance handling
• MinMax scaling for feature normalization
• GridSearchCV for hyperparameter optimization
• Cross-validation with F1-score optimization

📊 Model Evaluation Strategy

F1-score selected as primary metric - balances precision and recall for business impact where both false positives (unnecessary interventions) and false negatives (missed churners) carry costs.

---

🎯 Key Results

🏆 Best Model: Gradient Boosting

• F1-Score: 0.79 (2.8x better than baseline)
• Overall Accuracy: 89%
• Precision: 82% - High confidence in churn predictions
• Recall: 76% - Captures 3 out of 4 potential churners

📊 Model Comparison

Model	F1-Score	Accuracy	Business Value
Dummy Classifier	0.28	61%	Baseline
Logistic Regression	0.66	77%	Good interpretability
Random Forest	0.67	79%	Feature importance insights
Gradient Boosting	0.79	89%	Optimal performance
AdaBoost	0.74	86%	Strong alternative

🔍 Key Predictive Factors

1. Contract Type - Month-to-month customers at highest risk
2. Payment Method - Electronic check users show elevated churn
3. Service Tenure - Newer customers more likely to leave
4. Monthly Charges - Price sensitivity impacts retention
5. Internet Service Type - Fiber optic users require attention

---

💼 Business Impact

🎯 Immediate Actionable Insights

• Target month-to-month customers with loyalty programs
• Improve electronic check payment experience
• Focus retention efforts on fiber optic subscribers
• Implement graduated pricing for high-charge customers
• Enhance onboarding for new subscribers (first 90 days critical)

📈 Expected ROI

• 76% recall rate enables proactive intervention for majority of at-risk customers
• 82% precision minimizes wasted marketing spend on false positives
• Early intervention potential saves acquisition costs for retained customers

---

🛠️ Technical Implementation

Core Technologies

Python | Pandas | NumPy | Scikit-learn
Matplotlib | Seaborn | Jupyter Notebook

Advanced Techniques

SMOTE Oversampling | MinMax Scaling
Grid Search CV | Feature Engineering
Ensemble Methods | Statistical Analysis

---

📁 Project Structure

📦 Telecom_churn_predictor/
├── 📓 Telecom_churn_predictor.ipynb    # Complete analysis & methodology
├── 📊 data/
│   ├── contract.csv                    # Contract information
│   ├── personal.csv                    # Customer demographics  
│   ├── internet.csv                    # Internet service data
│   └── phone.csv                       # Phone service data
├── 📋 README.md                        # Project overview
└── 📄 requirements.txt                 # Environment dependencies

---

🚀 Quick Start

# Clone repository
git clone https://github.com/bergerache/Telecom_churn_predictor.git

# Install dependencies  
pip install pandas numpy matplotlib seaborn scikit-learn imbalanced-learn

# Launch analysis
jupyter notebook Telecom_churn_predictor.ipynb

---

🔮 Business Applications

Use Case	Implementation
🎯 Targeted Campaigns	Score customers monthly, prioritize top 20% risk
📞 Retention Calls	Automated alerts for high-risk customer segments
💰 Pricing Strategy	Adjust pricing based on churn probability scores
🎁 Promotional Offers	Customize incentives by risk factors and preferences

---

🏦 Banking & Fintech Applications

While this model was trained on telecom data, the methodology directly applies to financial services churn scenarios:

Telecom Scenario	Banking Equivalent
Customer cancels service	Account closure / Product attrition
Contract type impact	Fixed-term vs flexible savings
Payment method friction	Direct debit vs manual payment
Tenure analysis	Customer relationship length
Monthly charges sensitivity	Fee sensitivity / Price elasticity
Service usage patterns	Transaction frequency / Product utilisation

Direct Applications

• Deposit Account Churn — Predicting customers likely to close current/savings accounts
• Credit Card Attrition — Identifying cardholders at risk of cancellation
• Mortgage Refinancing — Flagging customers likely to switch lenders
• Investment Platform — Early warning for ISA/pension transfer risk

Why This Transfers

The feature engineering approach (tenure analysis, usage patterns, contract terms) translates directly to banking KPIs:

• Account age → Customer tenure
• Monthly charges → Fee revenue per customer
• Contract type → Product type (fixed/flexible)
• Payment method → Direct debit adoption
• Multiple services → Product holding depth

---

📖 Detailed Analysis

🔗 View Complete Jupyter Notebook

Comprehensive methodology, EDA insights, model development, and business recommendations

---

Built for data-driven customer retention

Transforming customer behavior analysis into strategic business value