Customer Churn Analysis & Prediction

Customer Churn Analysis & Prediction
Project Overview
Conducted an end-to-end Customer Churn Analysis project to identify patterns and predict customer attrition.
Focused on both data exploration (EDA) and machine learning modeling to generate actionable business insights.
Aimed to help businesses reduce churn, improve customer retention, and increase profitability.
Dataset Details
Dataset includes customer demographic, financial, and behavioral features:
RowNumber, CustomerId, Surname
CreditScore, Geography, Gender
Age, Tenure, Balance
NumOfProducts, HasCrCard, IsActiveMember
EstimatedSalary
Target Variable: Exited (Churn Status)
Tools & Technologies
Python Libraries:
pandas, numpy (data processing)
matplotlib, seaborn (data visualization)
scikit-learn (ML modeling)
Advanced Techniques:
SMOTE (handling class imbalance)
Feature scaling & encoding
Model evaluation metrics
Exploratory Data Analysis (EDA)
Performed deep EDA to uncover:
Customer behavior trends
Churn patterns across geography, age, and balance
Correlation between features and churn
Created visualizations:
Heatmaps, distributions, count plots
Identified key drivers of churn:
Age, inactivity, low engagement, and account balance
Machine Learning Models Implemented
Logistic Regression
Random Forest Classifier
K-Nearest Neighbors (KNN)
Support Vector Machine (SVM)
XGBoost
Gradient Boosting
Handling Imbalanced Data
Applied SMOTE (Synthetic Minority Oversampling Technique) to:
Balance churn vs non-churn classes
Improve recall and F1 score for minority class
Used class weighting for better model fairness
Model Performance Summary
Evaluated using:
Accuracy
Recall
F1 Score
ROC-AUC Score
Key Results:
Gradient Boosting
Best overall performer
Highest F1 Score: 0.598
Highest ROC-AUC: 0.859
XGBoost
Strong second-best model
Balanced precision & recall
Random Forest
High accuracy but weaker on churn detection
SVM & KNN
Moderate performance
Logistic Regression
Least effective for this dataset
Key Insights
Customer churn is strongly influenced by:
Low activity levels
Fewer product engagements
Demographic factors (age, geography)
Models like Gradient Boosting & XGBoost handle imbalance better and provide reliable predictions.
Business Impact
Helps businesses:
Predict high-risk customers
Design targeted retention strategies
Improve customer lifetime value (CLV)
Provides a data-driven foundation for decision-making
Deliverables
Cleaned and processed dataset
EDA report with visual insights
Trained ML models
Model comparison report
Prediction-ready pipeline