AI and technology

Supervised Machine Learning

A machine learning approach in which models learn from labeled examples (inputs paired with known outcomes) to predict outcomes for new data. In AML, labels usually come from past alert decisions and filed suspicious activity reports. It is effective for recurring patterns but limited in detecting risks that are absent from the training data.

Why it matters

Supervised models such as logistic regression, gradient-boosted trees and neural networks are widely used for fraud scoring, alert prioritization and customer risk rating. Because they learn directly from investigator decisions, they can help teams spend less time on alerts that closely resemble ones previously closed as false positives.

Their weakness is the labels themselves. Confirmed money laundering cases are rare, outcomes are often unknown, and the training data reflects only what earlier rules and investigators chose to look at. A model trained this way can reinforce existing blind spots and miss new typologies, which is why many programs pair supervised models with unsupervised anomaly detection.

Supervisors expect models used in AML to be explainable, independently validated and monitored for drift as customer behavior and criminal methods change. Investigators also need to understand why a model scored an alert as it did before acting on it.

Protect the flow. Power the growth.

See how ThetaRay turns regulatory expectations into detection and investigation that hold up to scrutiny.