Unsupervised learning

A type of machine learning in which algorithms identify patterns, structures, or relationships in data without being provided labeled examples or predefined correct outputs.

Source: AIEOG AI Lexicon (Feb 2026), adapted from NIST AI 100-1 and ISO/IEC 22989

Unsupervised learning is a way of training AI models where the algorithm explores data on its own, without being told what the “right answer” looks like. Instead of learning from labeled examples (this transaction is fraud, this one is not), the algorithm discovers hidden structures — groupings, anomalies, associations — that may not be obvious to human analysts.

Think of it as giving someone a pile of photographs with no labels and asking them to organize the photos into groups. They might sort by color, subject, location, or some other pattern they notice. Unsupervised learning works similarly: it finds structure in data without being told what structure to look for.

This makes unsupervised learning especially powerful for discovery — finding patterns that humans have not yet identified or would not think to look for.

Unsupervised learning is widely deployed across financial services, often in high-stakes applications:

• Anomaly detection in AML/BSA. Unsupervised models can identify unusual transaction patterns that do not match any known typology. This is critical because money laundering techniques constantly evolve, and rule-based systems only catch patterns they have been programmed to recognize.
• Customer segmentation. Banks and fintechs use unsupervised learning to group customers by behavior, risk profile, or product usage — informing everything from marketing strategy to risk-based monitoring.
• Fraud discovery. Unlike supervised fraud models that learn from historical fraud cases, unsupervised models can detect novel fraud schemes by identifying transactions that are statistically unusual.
• Market microstructure analysis. Trading firms use unsupervised learning to discover hidden patterns in order flow, price movements, and market behavior.
• Cyber threat detection. Unsupervised models identify unusual network activity, login patterns, or data access behavior that may indicate security breaches.

• Machine learning — the broader discipline encompassing unsupervised, supervised, and reinforcement learning
• Supervised learning — learning from labeled data with known correct outputs
• Semi-supervised learning — a hybrid approach using both labeled and unlabeled data
• Anomaly detection system — a common application of unsupervised learning
• Deep learning — neural network approaches that can be applied in unsupervised settings