📊 Telco Customer Churn Analysis - Complete Walkthrough

📊 Power BI Dashboard PDF:
You can download or view the full Power BI report here:

---

This project originally included detailed Python analytics and machine learning for customer churn prediction.
It has now been extended with a Power BI dashboard for deeper business insights and interactive visualization.

📋 Plot Views

📋 Project Overview

This project analyzes customer churn patterns for a telecommunications company using the Telco Customer Churn dataset from Kaggle. The goal is to understand churn drivers and build predictive models to identify at-risk customers.

Dataset Source: Kaggle - Telco Customer Churn
Original Data: IBM Sample Data Sets
Total Records: 7,043 customers
Features: 21 columns including demographic, service, and billing information

---

🎯 Business Context (Documented in Kaggle)

Predict customer behavior to develop focused retention programs by analyzing:

• Services each customer has signed up for
• Customer account information (tenure, contract, payment method)
• Demographic information
• Monthly and total charges

---

🔍 Data Understanding & Preprocessing

Initial Data Assessment

• Total Columns: 20 features + target variable (Churn)
• Missing Values: 11 records in TotalCharges & Tenure
• Data Type Issues: TotalCharges stored as object instead of numerical

Churn Distribution

• Overall Churn Rate: ~27% (imbalanced classification problem)
• Contract-wise Churn Analysis:

Contract Type	Total Customers	Churned	Not Churned	Churn Rate
Month-to-month	3,875	1,655	2,220	~43%
One year	1,473	166	1,307	~11%
Two year	1,695	48	1,647	~3%

Key Insight: Month-to-month contracts show significantly higher churn rates, indicating data leakage risk if not handled properly.

Missing Value Imputation Strategy

Root Cause Analysis: Missing values in TotalCharges occurred for customers with zero tenure (new customers not yet billed) - MNAR (Missing Not At Random) pattern.

Imputation Approach:

• Used median values due to high variability in monthly charges
• One year contracts: Median = $68.75
• Two year contracts: Median = $64.45
• Calculated TotalCharges = Tenure × MonthlyCharges for null records

Validation: Post-imputation scatter plots showed maintained data distribution integrity.

---

📊 Exploratory Data Analysis

Demographic Insights

Findings:

• Most customers are younger individuals without dependents
• Equal gender distribution
• Balanced marital status distribution

Financial Analysis by Contract Type

Key Discovery: Among month-to-month customers, churners consistently pay higher monthly charges than loyal customers. Even the cheapest 25% of churners (Q1) pay more than 50% of loyal customers.

Hypothesis Testing

Hypothesis 1: Additional Services Drive Churn

Assumption: Churners pay more due to additional services

Conclusion: ❌ Contradicted
Churners do have slightly more add-ons, but the difference isn't extreme enough to fully explain higher charges. Pricing structure appears reasonable.

Hypothesis 2: Short Tenure & Flexible Contracts Increase Churn

Assumption: Shorter tenure and month-to-month contracts lead to higher churn

Conclusion: ✅ Supported

• Churners have significantly shorter tenure
• Month-to-month contracts show highest churn probability
• Long-term contracts effectively reduce churn

Hypothesis 3: Payment Method Correlates with Churn

Assumption: Certain payment methods increase churn risk

Findings:

• Electronic check: Highest churn rate (53.7%)
• Bank transfer & credit card (automatic): Lower churn (~34%)
• Mailed check: Lowest churn (31.6%) but smaller user base

Conclusion: ❌ Partially Contradicted
Payment method alone doesn't fully explain churn patterns. Manual payments don't necessarily correlate with higher churn.

---

🤖 Machine Learning Modeling

Data Preparation

• Categorical Encoding: LabelEncoder for all categorical variables
• Class Imbalance Handling: SMOTE oversampling technique
• Train-Test Split: 70-30 split with stratification

SMOTE Validation

Validation Result: SMOTE maintains original data distribution while balancing classes, making it safe for modeling.

Model Comparison

Four models evaluated:

1. Logistic Regression
2. Random Forest
3. Gradient Boosting
4. XGBoost

Performance Metrics

Model	Accuracy	F1 Score	ROC-AUC
Logistic Regression	0.768	0.623	0.827
Random Forest	0.782	0.609	0.831
Gradient Boosting	0.774	0.619	0.836
XGBoost	0.777	0.617	0.834

Model Selection Rationale

Gradient Boosting Chosen because:

• Highest ROC-AUC (0.836) - best at ranking churn risk
• Maintains good F1 score balance
• Excellent for identifying high-risk customers for retention campaigns

Probability Analysis

Gradient Boosting Characteristics:

• Clear separation between low-risk (near 0.1) and high-risk (near 0.9) customers
• Confident predictions enable precise targeting
• Better risk ranking capability

Logistic Regression Characteristics:

• More conservative probability estimates
• Clustered around 0.2-0.8 range
• Less decisive for extreme cases

---

💡 Key Business Insights

1. High-Risk Customer Profile

• Contract: Month-to-month
• Tenure: Short-term (recent customers)
• Payment: Electronic check users
• Charges: Higher monthly billing tiers

2. Retention Opportunities

• Focus on month-to-month customers in early tenure period
• Encourage automatic payment methods
• Develop targeted offers for high-value at-risk customers

3. Model Deployment Value

Gradient Boosting provides clear customer segmentation:

• Low-risk (0.1 probability): Minimal retention effort needed
• High-risk (0.9 probability): Priority for retention campaigns
• Medium-risk: Monitor and nurture relationships

---

🚀 Recommendations

Immediate Actions

1. Targeted Campaigns: Focus retention efforts on month-to-month customers paying above median charges
2. Payment Incentives: Encourage automatic payment methods with small discounts
3. Early Intervention: Implement welcome programs for new month-to-month customers

Strategic Initiatives

1. Contract Benefits: Highlight advantages of longer contracts to month-to-month customers
2. Service Bundling: Create attractive bundles to increase value perception
3. Proactive Outreach: Use model predictions for pre-emptive retention efforts

Monitoring & Improvement

1. Model Retraining: Quarterly model updates with new data
2. Feature Enhancement: Incorporate customer service interaction data
3. A/B Testing: Validate retention strategy effectiveness

---

Report generated from comprehensive churn analysis workflow. Last updated: $(22-10-2025)

📁 Project Structure

├── Notebooks/01_null_value_imputation_walkthrough.ipynb
├── Notebooks/02_customer_churn_exploration_walkthrough.ipynb
├── Notebooks/03_EDA.ipynb
├── Notebooks/SmoteDataComparision.ipynb
├── Notebooks/Plots/ (All visualization assets)
└── Dataset/ (Processed data and model outputs)
└── Notebooks/ModelsProboScore/(Model outputs confidence on each prediction GB/LR)
---