🏦 Customer Churn Analytics Dashboard A Machine Learning–powered web application built using Python, Scikit-learn, and Streamlit to predict customer churn in European banking system based on customer demographics,financial behaviour, and product utilization patterns

Project Overview

• This project analyzes customer engagement,Product utilization,and churn behaviour using Machine Learning(XGBoost).
• The dashboard provides KPI insights, churn risk detection,and retention analysis.
• An interactive Streamlit dashboard was developed to visualize KPIs insights, predict churn risk detection and retention analysis.

Objectives

• Identify high-risk churn customers
• Analyze product utilization impact
• Measure customer retention strength
• Visualize feature importance
• Predict churn probability using Machine Learning

Problem Statement

Customer churn directly impacts revenue and long-term profitability. The goal of this project is to:

• Identify churn patterns
• Detect high-value at-risk customers
• Measure retention strength
• Build a predictive churn model

Dataset Description

The dataset contains customer information from a European bank. The following features were used for analysis and model development:

Column Description CreditScore - Customer creditworthiness

Geography - Customerlocation(France,Spain, Germany)

Gender - Male / Female

Age - Customer Age

Tenure - Number of Years with the bank

Balance - Account balance

NumOfProducts - Number of bank products

HasCrCard - Credit card ownership

IsActiveMember - Activity indicator

EstimatedSalary - Estimated annual salary

Exited - Churn indicator (target:1=churned,0=retained)

Dataset size :10,000 customers Features : 13 variables Target variable : Exited(Binary Classification)

Exploratory Data Analysis (EDA)

Exploratory Data Analysis (EDA) was performed to understand customer behavior, engagement levels, and factors influencing churn.

Key analyses performed:

• Customer churn distribution analysis
• Churn rate comparison across different geographies
• Product utilization vs churn analysis
• Active vs inactive member churn comparison
• Balance distribution analysis for churned customers
• Customer age distribution analysis
• Correlation heatmap to identify relationships between variables

Major Insights

Customers with low engagement and fewer banking products show higher churn probability, while active customers using multiple products demonstrate stronger retention.

Feature Selection

Final features used for modeling:

• CreditScore
• Geography
• Gender
• Age
• Tenure
• Balance
• NumOfProducts
• HasCrCard
• IsActiveMember
• EstimatedSalary

Dropped:

• CustomerId
• Surname

Machine Learning Models Tested

Three classification models were implemented and compared:

Model	Accuracy	ROC-AUC
Logistic Regression	~81%	~0.78
Random Forest	~84%	~0.85
XGBoost	~87%	~0.87

• final model selected : XGBoost
• Reason:
  • Higher accuracy
  • Better ROC-AUC curve
  • Improved recall for churn class
  • Strong feature importance interpretation

Machine Learning Model Evaluation Metrics

• Model Used:XGBoost Classifier
• Accuracy
• Precision
• Recall
• F1-Score
• ROC-AUC curve
• Confusion Matrix
• Feature Importance

Streamlit Dashboard Features

The interactive dashboard allows users to:

• Filter customers by geography
• Select age range
• Analyze churn distribution
• Detect high-value churn risks
• Analyze product utilization
• View model insights

Key UI Features

• Interactive sidebar filters
• Geography selection
• Age range slider
• Active member filter
• KPI metrics display
• Customer churn distribution visualization
• Product utilization analysis
• High-value churn risk indicators
• Clean and responsive dashboard layout

Project Workflow

1. Data Collection
2. Data Cleaning & Preprocessing
3. Exploratory Data Analysis
4. Feature Engineering
5. Model Training
6. Model Evaluation
7. Model Selection
8. Model Saving (Joblib)
9. Streamlit Dashboard Development
10. Deployment

Tech Stack

• Python
• Streamlit
• Pandas
• Numpy
• Matplotlib &Seaborn
• Scikit-Learn
• XGBoost

Project Structure

Customer_Churn_Analytics/ |

├── Streamlit_app_European_bank.py

├── Bank_analysis.ipynb

├── churn_model.pkl

├── best_model.pkl

├── European_Bank.csv

├── requirements.txt

└── README.md

Results & Conclusion

• The XGBoost model achieved the best performance among the tested models.
• Inactive customers show higher churn probability compared to active members.
• Customers using fewer banking products are more likely to churn.
• Product utilization and customer engagement significantly influence churn behavior.
• The Streamlit dashboard enables interactive analysis and churn risk identification.

Future Improvements

• Deploy the dashboard on cloud
• Add deep learning models
• Implement real-time churn prediction APIs
• Integrate advanced customer segmentation

Author

Jayanthi Varri

Developed using Python, Machine Learning and Streamlit