This project investigates how factors such as customer demographics, account activity, and financial behaviors influence churn risk at Deutsche Bank, a leading global financial institution. By analyzing relationships between these variables and churn outcomes, I built predictive models including Naive Bayes, Decision Trees, Random Forests, and XGBoost to forecast customer attrition and uncover retention opportunities. The objective is to deliver data-driven insights that optimize customer retention strategies and reduce financial losses. The project uses Python libraries like pandas, scikit-learn, XGBoost, and Matplotlib for data analysis, modeling, and visualization.
The Deutsche Bank Customer Churn Prediction project aims to:
- Identify Key Churn Drivers: Determine how customer age, account balance, credit score, and engagement affect churn likelihood
- Assess Data Patterns & Risk Factors: Detect trends, outliers, and attributes associated with high churn risk
- Build Predictive Models: Develop classification models (Naive Bayes, Decision Tree, Random Forest, and XGBoost) to forecast customer churn
- Support Proactive Retention Strategies: Recommend targeted interventions based on model insights to enhance customer loyalty
- Ensure Ethical & Reliable AI: Remove sensitive features (e.g., gender) to prevent bias and ensure fair modeling
The dataset contains 10,000 customer records from Deutsche Bank, each representing a unique customer and their banking details. Key features include:
- Exited: Indicates whether the customer churned (binary: 1 = churned, 0 = stayed)
- CreditScore: Customer’s credit score (continuous integer)
- Geography: Customer’s country (France, Spain, Germany)
- Age: Customer’s age in years (continuous integer)
- Tenure: Number of years as a bank customer (continuous integer)
- Balance: Account balance in euros (continuous float)
- NumOfProducts: Number of bank products used (discrete integer)
- HasCrCard: Whether the customer has a credit card (binary: 1 = yes, 0 = no)
- IsActiveMember: Whether the customer is an active member (binary: 1 = yes, 0 = no)
- EstimatedSalary: Estimated annual salary in euros (continuous float)
This dataset provides a comprehensive view of customer behavior and financial profiles, serving as a solid foundation for analyzing churn patterns and building predictive models to identify high-risk customers for Deutsche Bank.
- Confirmed no missing values or duplicates
- Removed irrelevant columns (
RowNumber,CustomerId,Surname) - Dropped gender to prevent bias in model predictions
- Encoded categorical variables (
Geography) via one-hot encoding
- Churn Distribution: 20% of customers churned, indicating moderate class imbalance
- Key Predictor Insights:
- Age: Customers >42.5 years old had significantly higher churn rates
- Balance: Customers with higher balances showed varied churn tendencies
- Active Membership: Inactive customers were 3x more likely to churn
- Engineered Feature: Created Loyalty (
Tenure/Age) to measure engagement over time
- CreditScore Impact: Customers with lower scores showed higher churn susceptibility
- Product Usage: Customers with >1 product were more likely to stay
- Geography: German customers had slightly higher churn rates than French/Spanish
- Served as an initial benchmark
- Accuracy: 80.6%
- Precision: 54.4%
- Recall: 30.3%
- F1-Score: 38.9%
- Highlighted limitations in handling mixed feature types
Four models were developed progressively for comparison:
- Naive Bayes (baseline)
- Decision Tree
- Random Forest
- XGBoost
Hyperparameter tuning via GridSearchCV with a stratified 75/25 train-test split ensured robust evaluation.
- Accuracy: 79.0%
- F1-Score: 56.1%
- Provided interpretable splits but risked overfitting
- Accuracy: 81.2%
- F1-Score: 58.0%
- Balanced predictive power and feature importance interpretability
- Accuracy: 81.8%
- F1-Score: 59.7%
- Emerged as the champion model due to:
- Highest predictive accuracy
- Strong handling of class imbalance
- Clear feature importance rankings
- Ensemble models (XGBoost, Random Forest) outperformed Naive Bayes by 20–21% in F1-score
- Feature Importance (XGBoost top predictors):
- EstimatedSalary
- Balance
- CreditScore
- Age
- Customers >42.5 years old with low activity were highest risk
- Engineered Loyalty feature improved model performance
- Removed sensitive variables (
Gender) to prevent bias - Applied stratified sampling to preserve class distribution
- Used F1-score as the primary metric to balance precision/recall
- Conducted cross-validation to ensure model reliability
- Age and activity status were the strongest churn predictors
- High-balance customers showed varied churn patterns, requiring targeted analysis
- German customers had slightly higher churn rates than other regions
| Model | Accuracy | F1-Score |
|---|---|---|
| Naive Bayes | 80.6% | 38.9% |
| Decision Tree | 79.0% | 56.1% |
| Random Forest | 81.2% | 58.0% |
| XGBoost | 81.8% | 59.7% |
XGBoost was selected as the champion model due to its superior predictive power and ability to handle class imbalance.
- End-to-End Churn Analysis: Identified key factors driving customer attrition
- High-Performance Predictive Modeling: Achieved 59.7% F1-score with XGBoost
- Actionable Business Insights:
- Target retention efforts for older, inactive customers
- Monitor high-balance customers for early churn signals
- Leverage Loyalty feature to identify at-risk long-term clients
- Ethical AI Practice:
- Excluded gender to prevent bias
- Ensured transparent, interpretable modeling
- Integrate churn predictions into CRM systems for real-time alerts
- Test personalized retention offers (e.g., fee waivers, loyalty bonuses)
- Expand feature engineering with transaction frequency and service usage data
- Conduct demographic fairness audits to ensure unbiased predictions
- Python: pandas, NumPy, scikit-learn, XGBoost
- Visualization: Matplotlib, Seaborn
- Model Evaluation: Precision-Recall, Confusion Matrices, Feature Importance
This solution equips Deutsche Bank with strategic, data-driven insights to proactively reduce customer churn, enhance retention strategies, and protect long-term revenue in a competitive banking landscape.