Machine Learning for Cryptocurrency AML Risk Analysis

Detecting and prioritising potentially illicit Bitcoin transactions using the Elliptic Bitcoin Dataset.

Project Overview

This project explores how machine learning can support the detection and prioritisation of potentially illicit cryptocurrency transactions in an anti-money-laundering and financial-intelligence context.

The objective is not to treat machine learning as proof of wrongdoing, but to examine how models can support screening, triage, and analytical prioritisation in large transaction environments.

Dataset

The project uses the Elliptic Bitcoin Dataset, a real-world dataset of Bitcoin transactions with engineered numerical features and transaction class labels.

The dataset includes:

• Illicit transactions

• Licit transactions

• Unknown transactions

The dataset is not included in this repository. It should be downloaded from its original source.

Methods Used

• Principal Component Analysis

• K-means clustering

• Linear regression

• Ridge regression

• Lasso regression

• Random Forest regression

• Logistic regression

• Support Vector Machine

• Random Forest classification

Key Results

The Random Forest classifier achieved the strongest classification performance on the labelled subset:

| Metric | Random Forest |

| Accuracy | 98.68% |

| Precision | 99.37% |

| Recall | 87.02% |

| F1-score | 92.79% |

The model correctly identified a high proportion of illicit transactions while maintaining a very low false-positive rate.

Selected Visuals

Classification Model Comparison

Random Forest Confusion Matrix

Feature Importance

Repository Structure

crypto-aml-machine-learning/

│

├── README.md

├── LICENSE

├── .gitignore

│

├── report/

│   └── Alejandro_Soba_Crypto_AML_ML_Report.docx

│

├── src/

│   └── crypto_aml_ml_analysis.qmd

│

└── figures/

    ├── 01_class_distribution.png

    ├── 02_pca_projection.png

    ├── 03_clustering_visual.png

    ├── 04_regression_model_comparison.png

    ├── 05_classification_model_comparison.png

    ├── 06_confusion_matrix_rf.png

    └── 07_feature_importance.png