Can Machine Learning Catch Criminals Before the Blockchain Does? – Report

Traditional AML tools struggle to keep pace with the scale and speed of illicit activity in crypto. In this exploratory study, the AMLBot Team investigates whether behavior-based machine learning can spot suspicious entities before they appear on blacklists and how explainability tools like SHAP make this process transparent.

We tested over 1,000+ behavioral features, built a classification model using gradient boosting, and evaluated it on millions of real clusters. The model is currently in internal testing, with results informing our roadmap toward production use.

Read the full research paper to explore our findings, SHAP plots, and what this means for the future of crypto compliance.