-61DAFB){kind=icon}
An end-to-end AI-powered fraud detection system designed to identify fraudulent credit card transactions while providing clear, human-understandable explanations for every prediction. The system combines a high-performance machine learning model with Explainable AI (XAI) techniques and is deployed as a full-stack web application.
Credit card fraud detection is a high-stakes problem where incorrect decisions can lead to financial loss or customer dissatisfaction. This project addresses that challenge by not only detecting fraudulent transactions with high accuracy but also explaining why a transaction was classified as fraud.
The solution uses XGBoost for robust prediction, SHAP for explainability, and a Flask + React (Vite) stack for deployment, making the system practical, transparent, and user-friendly.
This project uses the Credit Card Fraud Detection Dataset from Kaggle.
🔗 Dataset Link:
https://www.kaggle.com/datasets/mlg-ulb/creditcardfraud
Dataset highlights:
- Real European credit card transaction data
- Highly imbalanced dataset (fraud < 1%)
- Features V1–V28 are PCA-transformed to protect confidentiality
- Class is the target variable:
0→ Legitimate transaction1→ Fraudulent transaction
The following preprocessing steps were applied to prepare the data for modeling:
- Dropped the Time feature as it did not contribute significantly to prediction
- Standardized the Amount feature using
StandardScaler - Conducted Exploratory Data Analysis (EDA) to analyze fraud distribution and feature behavior
- Applied SMOTE to handle severe class imbalance
- Split the dataset into training and testing sets
XGBoost was selected due to its strong performance on structured/tabular data and its ability to handle class imbalance effectively.
Why XGBoost?
- Excellent performance on imbalanced datasets
- Captures complex, non-linear relationships
- High predictive accuracy and ROC-AUC scores
- Efficient and scalable
To avoid black-box predictions, SHAP (SHapley Additive exPlanations) was integrated into the system.
SHAP enables:
- Feature-level contribution analysis for each prediction
- Local and global interpretability
- Visual explanations showing why a transaction was flagged as fraud
This improves trust, transparency, and accountability—critical factors in financial applications.
The model was evaluated using multiple metrics:
- Accuracy: ~99.5%
- Precision: ~91%
- Recall: ~82%
- F1-Score
- ROC-AUC Score
- Confusion Matrix
Given the imbalanced nature of the dataset, Recall and ROC-AUC were prioritized over raw accuracy.
The trained model was exported and integrated into a full-stack web application:
- Backend: Flask REST API for model inference and SHAP explanation generation
- Frontend: React (Vite) for an interactive and responsive UI
- Model Integration: Real-time predictions with explainability
- Packaging: The complete application was packaged and delivered as an APK
This ensures smooth end-to-end functionality—from transaction input to prediction and explanation.
- Python
- Pandas, NumPy
- Scikit-learn
- XGBoost
- SHAP
- Imbalanced-Learn (SMOTE)
- Flask (Backend / REST API)
- React (Vite) (Frontend)
- Model Serialization
- RESTful APIs
This project was developed as a collaborative team effort. I was primarily responsible for the deployment and full-stack implementation of the system. My contributions included exporting and optimizing the trained XGBoost model, developing the Flask-based backend API, and building the frontend interface using React (Vite).
I handled the end-to-end model integration, enabling real-time fraud prediction and SHAP-based explainability within the web application. Additionally, I packaged the complete solution as an APK, ensuring seamless usability from model inference to user interaction.
- High-performance fraud detection with explainability
- Real-world imbalanced dataset handling
- SHAP-based transparent AI decisions
- Full-stack deployment with modern web technologies
- End-to-end system from ML to APK delivery
⭐ If you find this project useful, feel free to star the repository!