Fraud Detection System

This project implements a comprehensive fraud detection system using multiple machine learning algorithms and advanced feature engineering techniques.

Features

• Data preprocessing and cleaning
• Advanced feature engineering
• Multiple ML models (Random Forest, XGBoost, Logistic Regression, Isolation Forest)
• Ensemble model with meta learner
• Comprehensive evaluation metrics
• Class imbalance handling

Setup Instructions

1. Clone the Repository

   git clone <repository-url>
   cd <repository-directory>
   

2. Install Dependencies

   pip install -r requirements.txt
   

3. Run the Fraud Detection Analysis Use the run script:

   python run_fraud_detection.py
   

   Or, run with a custom sample size:

   python run_with_sample.py <sample_size>
   

Results (1,000,000 Samples)

• Random Forest: Accuracy 100.00%, F1-Score 99.91%, ROC-AUC 99.95%
• XGBoost: Accuracy 100.00%, F1-Score 99.85%, ROC-AUC 99.98%
• Logistic Regression: Accuracy 99.99%, F1-Score 99.60%, ROC-AUC 99.98% (Best Performer)
• Meta Ensemble: Accuracy 100.00%, F1-Score 99.85%, ROC-AUC 99.91%

Top Important Features

1. amount_equals_old_balance (28.54%)
2. step (25.51%)
3. balance_diff_orig (11.74%)

Visualization

• Evaluation plots and feature importance plots have been saved:
  • fraud_detection_evaluation.png
  • feature_importance.png

Meta Learner Approach

Utilizes a Gradient Boosting Classifier to refine ensemble predictions for greater accuracy. The meta learner learns from ensemble predictions to provide a final probability score.