🛡️ FraudShield — Credit Card Fraud Detection System

A machine learning-based fraud detection system designed to identify anomalous financial transactions using both supervised and unsupervised learning approaches. The project includes model training, evaluation, explainability, and a live Streamlit dashboard for real-time fraud prediction.

---

🚀 Overview

Fraud detection is a highly imbalanced classification problem where fraudulent transactions represent a very small portion of total activity (~0.17%). This project aims to build a robust system capable of detecting fraud with high recall while maintaining precision.

---

📊 Dataset

• Source: Kaggle Credit Card Fraud Dataset
• Download creditcard.csv from Kaggle and place it inside the data/ folder.
• Total Records: 284,807
• Fraud Cases: 492 (~0.17%)
• Features:
  • V1–V28: PCA-transformed anonymized features
  • Time, Amount
  • Target: Class (0 = Normal, 1 = Fraud)

---

🧱 Project Architecture

ALLANAI.FraudShield/ ├── dashboard/ │ └── app.py # Streamlit dashboard ├── models/ │ ├── xgboost_fraud_model.pkl │ └── scaler.pkl ├── notebooks/ │ └── 01_fraudshield_eda_model.ipynb ├── requirements.txt └── README.md

---

🔍 Exploratory Data Analysis (EDA)

Key insights:

• Dataset is highly imbalanced (~0.17% fraud)
• Most transactions are low-value with long-tail distribution
• Fraud patterns are not linearly separable
• PCA features require non-linear models

---

🤖 Models Implemented

1. XGBoost (Supervised Learning)

• Handles imbalanced data using scale_pos_weight
• Captures complex non-linear relationships

Performance:

• Precision: 0.71
• Recall: 0.87
• F1 Score: 0.78

---

2. Isolation Forest (Unsupervised Learning)

• Detects anomalies without labelled data
• Useful as baseline anomaly detector

Performance:

• Precision: 0.29
• Recall: 0.30
• F1 Score: 0.30

---

⚖️ Model Comparison

Model	Precision	Recall	F1 Score
XGBoost	0.71	0.87	0.78
Isolation Forest	0.29	0.30	0.30

Key Insight

Supervised learning significantly outperforms unsupervised anomaly detection in this dataset due to availability of labelled fraud data.

---

🔍 Model Explainability (SHAP)

SHAP (SHapley Additive exPlanations) is used to interpret model predictions by identifying the contribution of each feature.

This improves:

• Transparency
• Trust in predictions
• Debugging capability

---

🖥️ Streamlit Dashboard

An interactive dashboard allows real-time fraud prediction.

Features:

• Input transaction details
• Predict fraud probability
• Display risk level
• Show model performance metrics

Run locally:

streamlit run dashboard/app.py