Project Overview

This project implements a machine learning solution to flag fraudulent transactions given the constraints of the business problem: aka, only 400 transactions can be reviewed per month, and the goal is to prevent the most amount of fraud value.

Key Features:

• Time-based train-test split to simulate production conditions (last month used as test data)
• Custom feature engineering for fraud detection
• Machine learning model (XGBoost) for fraud prediction
• Rule-based fraud detection for comparison
• Cross-validation for robust model evaluation
• Performance evaluation using a custom fraud capture score

Project Structure

├── data/
│   ├── transactions_obf.csv
│   └── labels_obf.csv
├── src/
│   ├── main.py
│   ├── data_loader.py
│   ├── feature_engineering.py
│   ├── model_pipeline.py
│   └── evaluation.py
└── README.md

Installation and Setup

1. Install required packages ussing requirements.txt:

   pip install -r requirements.txt
   

   Alternatively, if there are any clashes, the following should work:

   pip install pandas numpy scikit-learn xgboost imbalanced-learn joblib
   

2. Ensure the data files are in the data/ directory

Usage

Run the main script:

python src/main.py

This will:

1. Load and preprocess the data
2. Perform cross-validation
3. Train the final model
4. Evaluate the model on the test set
5. Compare the ML model with a rule-based approach
6. Save the trained model

Model Performance

The machine learning model significantly outperforms the rule-based approach:

• ML Model Fraud Capture Score: 90.52%
• Rule-Based Model Fraud Capture Score: 63.93%
• Improvement: 26.59%

Cross-validation results:

• ML Model Average Fraud Capture Score: 88.36% (+/- 3.89%)
• Rule-Based Model Average Fraud Capture Score: 77.51% (+/- 8.16%)
• Average Improvement: 10.86%

Key Components

• DataLoader: Loads transaction and label data
• DataPreprocessor: Performs initial data preprocessing
• CustomFeatureEngineer: Implements domain-specific feature engineering
• FraudDetectionPipeline: Combines feature engineering, preprocessing, and model training
• evaluate_models_with_cv: Performs cross-validation for model evaluation
• fraud_capture_score: Custom metric for evaluating model performance

Future Improvements

1. Feature importance analysis for better understanding of the model
2. Hyperparameter tuning for potentially improved performance
3. Exploration of other machine learning algorithms