Anti-Money Laundering Detection System

Production-Ready AML Detection for Banking Compliance

Project Overview

A machine learning-powered Anti-Money Laundering (AML) detection system with an interactive Streamlit dashboard. Built using IBM's Synthetic AML Dataset (31.9M transactions), this system achieves 73% precision and ROC-AUC 97.8%, making it production-ready for banking compliance teams.

Live Demo

Visit Live Dashboard

Key Features

• Real-time Transaction Scoring: Instant risk assessment for individual transactions
• Calibrated Risk Scores: Reliable probability estimates (when model says 10% risk, it means 10%)
• Automatic Feature Engineering: Extracts 20+ risk indicators from raw transaction data
• Production-Optimized: 27x reduction in false positives vs baseline models
• Interactive Dashboard: User-friendly interface for compliance analysts

Model Performance

• Precision: 72.75% (73% of alerts are true laundering cases)
• Recall: 18.87% (catches 1 in 5 laundering cases)
• ROC-AUC: 97.81% (excellent discrimination)
• False Positive Rate: 0.01% (only 752 false alarms out of 6.4M normal transactions)
• Daily Alert Volume: ~860 alerts (manageable workload)

Project Structure

├── app/                    # Streamlit dashboard
│   ├── Home.py            # Main page with KPIs and model overview
│   ├── pages/
│   │   ├── 01_Model_Validation.py      # Performance metrics & confusion matrix
│   │   ├── 02_Investigator_Workbench.py # Real-time transaction scoring
│   │   └── 04_Data_Insight.py          # EDA visualizations
├── models/                 # Trained model artifacts
│   ├── calibrated_lightgbm_model.pkl
│   ├── scaler.pkl
│   ├── feature_names.json
│   └── model_config.json
├── note/              # Jupyter notebooks for model development
├── data/                   # find it on IBM website
└── requirements.txt        # Python dependencies

Tech Stack

• ML Framework: LightGBM with Isotonic Calibration
• Dashboard: Streamlit
• Data Processing: Pandas, NumPy
• Visualization: Plotly
• Deployment: Aws is cooming soon

Dataset

• Source: IBM Synthetic AML Dataset
• Size: 31,898,238 transactions
• Period: September 1-28, 2022
• Laundering Cases: 35,230 (0.11%)
• Class Imbalance: 1:905 ratio

Team Members

Catholic University of America - MDA Capstone Project Fall 2025

• Delphin Kaduli
• Tycho Janssen
• Solomon Pinto

Quick Start

Prerequisites

• Python 3.8+
• pip

Installation

# Clone the repository
git clone https://github.com/DelphinKdl/CUA-MDA-Capstone-BaaS-Risk-Monitoring.git
cd CUA-MDA-Capstone-BaaS-Risk-Monitoring

# Create virtual environment
python -m venv .venv

# Activate virtual environment
# On macOS/Linux:
source .venv/bin/activate
# On Windows:
.\.venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Run the dashboard
streamlit run app/Home.py

The dashboard will open in your browser at http://localhost:8501

Dashboard Pages

1. Home

• Key performance indicators (KPIs)
• Champion model overview
• Business impact metrics

2. Model Validation

• Confusion matrix visualization
• Performance metrics
• Model configuration details

3. Investigator Workbench

• Star Feature: Real-time transaction risk scoring
• User-friendly input form
• Automatic feature engineering
• Instant risk assessment with recommendations

4. Data Insights

• Dataset overview and statistics
• Class distribution analysis
• Payment format patterns
• Temporal patterns (day of week)
• Structuring detection analysis

Key Risk Indicators Detected

• ACH Payment Format: 49x baseline risk
• Weekend Transactions: 3x baseline risk
• Structuring Pattern: $9K-$10K range (8.3x risk)
• High-Risk Institutions: Bank-level risk patterns
• Currency Risk Patterns: UK Pound structuring
• Temporal Patterns: Day of week and hour analysis

Model Details

• Algorithm: LightGBM with Probability Calibration
• Training Data: 25.5M transactions (80%)
• Test Data: 6.4M transactions (20%)
• Features: 20 engineered features
• Threshold: 0.10 (optimized for precision)
• Training Strategy: Full dataset (no undersampling)

Business Impact

• 62% reduction in analyst workload vs alternative models
• 27x reduction in false positives
• 73% confidence in every alert
• Manageable alert volume: ~860 alerts per day
• Precision-focused strategy: Maximizes analyst efficiency

Acknowledgments

• IBM for providing the Synthetic AML Dataset
• Catholic University of America MDA Program
• Project advisors: Dr. Matthew Jacob

License

MIT License

Copyright (c) 2025 Delphin Kaduli, Tycho Janssen and Solomon Pinto.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights