This project predicts whether a customer will default on their credit card payment using machine learning techniques.
- Predict default risk (High / Low)
- User-friendly web interface (Flask)
- Uses real financial dataset
├── app.py
├── model.pkl
├── scaler.pkl
├── train_model.py
├── templates/
│ └── index.html
├── data/
│ └── default of credit card clients.xls
└── requirements.txt
- Python
- Pandas, NumPy
- Scikit-learn
- Flask
This project uses the Default of Credit Card Clients dataset from the UCI Machine Learning Repository.
- Source: https://archive.ics.uci.edu/dataset/350/default+of+credit+card+clients
- Records: 30,000 customers
- Features: 23 input variables
- Target: Default payment next month (Yes/No)
The dataset contains customer demographic information, credit limit, repayment history, bill statements, and payment records to predict credit card default risk.
-
Clone the repository: git clone https://github.com/jincyjos22/Credit-Card-Default-Prediction-Using-Machine-Learning.git
-
Go to project folder: cd Credit-Card-Default-Prediction-Using-Machine-Learning
-
Install dependencies: pip install -r requirements.txt
-
Run model training: python train_model.py
-
Run the app: python app.py
-
Open browser: http://192.168.1.6:5000/ http://127.0.0.1:5000
- High Risk → Customer likely to default
- Low Risk → Customer not likely to default
- 🟢 Green result → Low Risk (Customer is likely to repay)
- 🔴 Red result → High Risk (Customer may default)
The color coding helps users quickly understand the prediction result.
Machine Learning based Credit Card Default Prediction
Credit card default is a major problem for financial institutions, leading to financial losses and increased risk.
This project uses machine learning techniques to predict whether a customer will default on their credit card payment based on financial and repayment history data. The goal is to help banks identify high-risk customers early and improve credit decision-making.
The primary goal of the research is the development and assessment of several machine learning models which could forecast credit card customer default.
Moreover, the project should involve comparison of algorithms and determination of the most accurate model.
The dataset used in this project is obtained from the UCI Machine Learning Repository, titled “Default of Credit Card Clients Dataset.” It is available as an Excel file and contains information about credit card clients in Taiwan.
The dataset includes the following features:
- LIMIT_BAL → Credit limit amount of the customer
- SEX → Gender (1 = Male, 2 = Female)
- EDUCATION → Education level (1 = Graduate School, 2 = University, 3 = High School, 4 = Others)
- MARRIAGE → Marital status (1 = Married, 2 = Single, 3 = Others)
- AGE → Age of the customer
- PAY_0 → Repayment status in September
- PAY_2 → Repayment status in August
- PAY_3 → Repayment status in July
- PAY_4 → Repayment status in June
- PAY_5 → Repayment status in May
- PAY_6 → Repayment status in April
- BILL_AMT1 → Bill amount in September
- BILL_AMT2 → Bill amount in August
- BILL_AMT3 → Bill amount in July
- BILL_AMT4 → Bill amount in June
- BILL_AMT5 → Bill amount in May
- BILL_AMT6 → Bill amount in April
- PAY_AMT1 → Payment made in September
- PAY_AMT2 → Payment made in August
- PAY_AMT3 → Payment made in July
- PAY_AMT4 → Payment made in June
- PAY_AMT5 → Payment made in May
- PAY_AMT6 → Payment made in April
-
default.payment.next.month
- 1 → Customer will default (fail to pay)
- 0 → Customer will not default
Data preprocessing was performed to clean and prepare the dataset for machine learning.
● Missing Values
The dataset was checked for missing values. Since very few or none were present, median imputation was used for numerical features.
● Outliers
Outliers in features like LIMIT_BAL and bill amounts were detected using boxplots and treated where necessary.
● Duplicate Records
Duplicate entries were removed to ensure data quality and avoid bias in training.
● Categorical Variables
Categorical features (SEX, EDUCATION, MARRIAGE) were already numerically encoded, so no additional encoding was required.
● Skewed Data
Skewed numerical features were analyzed, and scaling was applied to improve model performance.
● Feature Engineering
A new feature TOTAL_PAY was created by summing PAY_AMT1 to PAY_AMT6 to capture overall payment behavior.
● Feature Scaling
StandardScaler was applied to normalize numerical features for better model performance.
This project includes a set of simple and effective visualizations to understand customer behavior and factors influencing credit card default.
A count plot is used to visualize the number of customers who defaulted vs those who did not. This helps in understanding the class distribution of the dataset.
A histogram with KDE (Kernel Density Estimation) is used to observe the distribution of customer age.
This helps identify the most common age groups in the dataset.
A box plot is used to compare credit limits between defaulters and non-defaulters.
This helps identify whether lower or higher credit limits are associated with default risk.
A bar plot is used to analyze the relationship between repayment status (PAY_0) and default probability.
This is one of the most important visualizations, showing that delayed payments strongly increase the risk of default.
A KDE plot is used to visualize the distribution of total payments made by customers over the last six months.
This helps understand general payment behavior.
A heatmap is used to show correlations between key features such as credit limit, age, payment delay, and total payment.
This helps identify relationships between variables and their impact on default.
The visualizations highlight that:
- Customers with delayed payments are more likely to default
- Payment behavior plays a critical role in predicting default
- Credit-related features also influence risk
These insights are used to support the machine learning model for predicting credit card default.
Feature engineering involves transforming and creating new features to improve the performance of machine learning models.
In this dataset, categorical variables such as SEX, EDUCATION, and MARRIAGE are already represented in numerical form. Therefore, additional encoding techniques like One-Hot Encoding or Label Encoding were not required.
However, these variables were carefully reviewed to ensure consistency and correctness before model training.
A new feature called TOTAL_PAY was created by summing all payment amounts from the past six months (PAY_AMT1 to PAY_AMT6).
[ \text{TOTAL_PAY} = \text{PAY_AMT1} + \text{PAY_AMT2} + \text{PAY_AMT3} + \text{PAY_AMT4} + \text{PAY_AMT5} + \text{PAY_AMT6} ]
This feature captures the overall payment behavior of customers and helps the model better understand their repayment capacity.
Important features were identified based on correlation analysis and domain knowledge. Repayment history variables (PAY_0 to PAY_6) were found to be highly influential in predicting default.
The Credit Card Default dataset contains financial and behavioral features with significantly different value ranges. For example:
Credit limit and bill amounts are very large values Payment amounts vary widely Age is within a small range Repayment status values are small integers
To ensure that all features contribute equally during model training, feature scaling was applied.
Technique Used
StandardScaler (Z-score normalization) was used to standardize all numerical features.
Reason for Selection Financial features like bill amounts and credit limits have large variations Scaling prevents these features from dominating smaller ones like age or repayment status Essential for algorithms such as SVM and KNN Conclusion
Feature scaling improved model performance by normalizing all financial and behavioral features in the dataset.
The goal of this project is to predict whether a customer will default on their credit card payment based on historical financial data.
Dataset Handling Input features include customer demographics, credit limit, repayment history, bill statements, and previous payments Target variable: Default payment next month (1 = Yes, 0 = No)
The dataset was split into:
Training data (80%) Testing data (20%) Models Applied Logistic Regression – baseline classification model Decision Tree – captures decision rules from financial behavior Random Forest – improves prediction by combining multiple trees Support Vector Machine (SVM) – effective for classification boundaries K-Nearest Neighbors (KNN) – uses similarity between customers Conclusion
Multiple models were trained to understand different patterns in customer behavior and identify the most accurate model for predicting default risk.
Models were evaluated to identify customers likely to default.
📊 Evaluation Metrics Accuracy – overall correctness Precision – correctness of predicted defaulters Recall – ability to detect actual defaulters F1-Score – balance between precision and recall Confusion Matrix – prediction breakdown ROC-AUC – model discrimination ability
Key Insight
The dataset is imbalanced, so Recall is more important than Accuracy, as missing defaulters leads to financial risk.
🏆 Best Model
Random Forest performed best due to:
Better handling of complex patterns Reduced overfitting Stable performance across metrics 📌 Conclusion
Model evaluation helped select the most reliable model for predicting credit card default risk.
Hyperparameter tuning was performed to improve the performance of the Random Forest model.
- Used systematic search (GridSearchCV)
- Optimized number of trees, max depth, and split criteria
- Reduced overfitting and improved generalization
Improved model accuracy and performance on unseen data.
The trained model is deployed using a Flask web application.
- Model saved using
joblib - Flask handles user input and prediction
- Users enter financial details through a web form
- 🟢 Low Risk → Customer is likely to repay
- 🔴 High Risk → Customer may default
This project successfully developed a machine learning model to predict credit card default risk using real-world financial data.
- Customers with delayed repayment history are more likely to default
- Higher outstanding bills increase default risk
- Payment behavior is a strong indicator of future default
- Random Forest provided the most accurate and reliable predictions for this dataset.
- Helps banks identify high-risk customers
- Supports credit approval decisions
- Reduces financial losses
- Use larger and more recent financial datasets
- Apply advanced models like deep learning
- Improve feature selection and engineering
- Deploy as a scalable production web application