💳 Bank Customer Churn Prediction – Artificial Neural Network (ANN)

Quick Snapshot:

Predict bank customer churn using an Artificial Neural Network (ANN). Identify high-risk customers, evaluate model performance, and visualize predictions for actionable insights.

📌 Project Overview

This project applies supervised machine learning to predict whether a bank customer will stay or churn, based on profile and transactional features. By identifying customers likely to leave, banks can design targeted retention strategies.

---

🎯 Background

Customer churn impacts profitability and long-term growth in banking. Proactive churn prediction allows banks to:

• Retain high-value customers
• Reduce attrition-related losses
• Improve customer experience through tailored interventions

---

🗂️ Dataset Overview

Source: Provided by Data Science Bootcamp (10,000 records)

Features Used for ANN Prediction:

Feature	Description
Geography	Customer country (France, Spain, Germany)
Credit Score	Customer credit score (300–850)
Gender	Male or Female
Age	Customer age in years
Tenure	Number of years as a bank customer
Balance	Account balance in USD
Number of Products	Total bank products held by the customer
Has Credit Card	1 if customer has a credit card, 0 otherwise
Is Active Member	1 if active, 0 otherwise
Estimated Salary	Annual estimated salary in USD

Problem Type: Supervised learning – Classification (Predicting churn: STAY vs CHURN)

⚠️ Disclaimer Dataset is synthetic/educational; any real-world conclusions should not be drawn from this data.

---

🛠️ Tools I Used

• Python (NumPy, Pandas, Matplotlib, Seaborn)
• Scikit-learn (MLPClassifier, preprocessing, metrics)
• Jupyter Notebook for experimentation and visualization

---

🚀 Getting Started

Prerequisites

• Python 3.11 or higher
• Jupyter Notebook or Google Colab
• Conda (recommended) or pip

Installation

1. Clone the repository

   git clone https://github.com/NadiaRozman/Bank_Customer_Churn_ANN.git

2. Install required packages (if not already installed):

   pip install numpy pandas matplotlib seaborn scikit-learn

3. Open and run the notebook

   • Navigate to Bank_Customer_Churn_ANN.ipynb
   • Ensure bank_customer_churn_data.csv is in the correct directory
   • Run all cells to reproduce the analysis

---

🔬 ANN Model Implementation

1. Preprocessing:

• Label Encoding, One-Hot Encoding, Feature Scaling

2. ANN Architecture:

• Input Layer: 12 features
• Hidden Layer: 100 neurons, ReLU activation
• Output Layer: 1 neuron, Sigmoid (for binary classification)
• Optimizer: Adam
• Loss Function: Cross-Entropy
• Max Iterations: 200

3. Model Evaluation:

• Training Accuracy: 87.90%
• Test Accuracy: 86.25%
• AUC Score: 0.852 (Very good performance)

Confusion Matrix

Figure 1: Confusion Matrix showing model predictions vs actual labels for stayed vs churned customers.

ROC Curve

Figure 2: ROC Curve of ANN Customer Churn Model. Shows trade-off between true positive rate and false positive rate. AUC = 0.852.

---

📊 Customer Predictions

Single Customer Example

Customer Profile:

• Geography: France
• Credit Score: 600
• Gender: Male
• Age: 40
• Tenure: 3
• Balance: $60,000
• Number of Products: 2
• Has Credit Card: Yes
• Is Active Member: Yes
• Estimated Salary: $50,000

Prediction:

Metric	Value
Churn Probability	2.53%
Risk Level	LOW RISK
Recommendation	Maintain standard service

Multiple Customer Evaluation

Example Output:

Customer	Churn Probability	Risk Level	Prediction
Customer 1	2.53%	LOW	STAY
Customer 2	52.29%	HIGH	CHURN
Customer 3	94.82%	HIGH	CHURN
Customer 4	99.92%	HIGH	CHURN
Customer 5	8.15%	LOW	STAY

Summary Statistics:

• Total Customers Evaluated: 5
• High Risk (>50%): 3
• Medium Risk (30–50%): 0
• Low Risk (<30%): 2
• Average Churn Probability: 51.54%

Churn Probability Visualization

Figure 3: Combined visualization showing churn probability per customer (bar chart) and risk level distribution (pie chart). Colors indicate risk levels: Low (green), Medium (orange), High (red).

---

💡 Insights & Business Implications

• Customers with low credit scores, low balances, or inactive status are more likely to churn.
• High churn probability among some long-term customers may indicate service dissatisfaction or competitive offers.
• Bank can prioritize retention campaigns for HIGH risk customers to reduce overall churn.
• Risk-based segmentation allows targeted incentives, improving ROI of retention strategies.

---

📚 What I Learned

• ANN can handle structured tabular data effectively for classification tasks.
• Feature scaling and encoding are crucial for ANN performance.
• Visualization of predictions aids stakeholders in understanding model insights.
• Interpreting model outputs in a business context is key for actionable recommendations.

---

🔮 Future Enhancements

• Explore hyperparameter tuning to improve ANN performance and convergence.
• Implement cross-validation for more robust evaluation.
• Integrate additional customer features like transaction history, complaints, and product usage.
• Deploy model into a dashboard for live churn monitoring and decision-making.
• Compare ANN results with other classifiers (Random Forest, XGBoost) for benchmarking.

---

✨ Created by Nadia Rozman | January 2026

📂 Project Structure

ANN_Bank_Customer_Churn_Prediction/
│
├── data/
│   └── bank_customer_churn_data.csv
│
├── images/
│   ├── 1_confusion_matrix.png
│   ├── 2_roc_curve.png
│   └── 3_multiple_customers_prediction.png
│
├── notebook/
│   └── Bank_Customer_Churn_ANN.ipynb
│
└── README.md

🔗 Connect with me

• GitHub: @NadiaRozman
• LinkedIn: Nadia Rozman

⭐ If you found this project helpful, please consider giving it a star!