This repository contains a full machine learning pipeline to detect fraudulent credit card transactions. It combines geospatial, temporal, and behavioral features with ensemble learning methods to improve detection performance.
- Feature Engineering: Age, transaction time, day of week, and distance traveled.
- Geospatial Analysis: Calculates distances between customer and merchant locations using latitude and longitude.
- Suspicious Travel Detection: Flags implausible high-speed travel between transactions.
- Imbalanced Data Handling: Uses SMOTE to address severe class imbalance.
- Modeling: Random Forest and XGBoost for robust fraud classification.
- Python
- Libraries:
pandas,numpy,scikit-learn,matplotlib,seaborn,geopy,xgboost,imbalanced-learn
The data used for this project is stored locally and includes columns such as:
- Transaction time and date
- Customer and merchant geolocation
- Transaction amount and category
- Demographics (e.g., gender, date of birth)
- Link to dataset: https://www.kaggle.com/datasets/kelvinkelue/credit-card-fraud-prediction
- Gender-based transaction distributions
- Day-of-week transaction trends
- Correlation matrix of engineered features
- Clone the repo:
git clone https://github.com/your-username/credit-card-fraud-detection.git
cd credit-card-fraud-detection- Install dependencies:
pip install -r requirements.txt- Run the notebook or script to train and evaluate the model.
- Precision
- Recall
- F1 Score
- ROC-AUC
The final model is evaluated using an optimized threshold derived from the precision-recall curve. A classification report is printed to assess its performance.
Created by Ruwaida Al Harrasi – feel free to reach out!
Feel free to fork the repository or suggest improvements via pull requests!