Unsupervised Machine Learning

Find the fraud you have never seen before

Detect anomalous, coordinated, and emerging behavior without waiting for historical fraud labels. DataVisor's patented Unsupervised Machine Learning learns from the population itself to surface patterns that rules and labeled models have not been taught to recognize.

Request a Demo →
Fraud & Risk Platform/Unsupervised ML
Solution Pillars

Learn the pattern from the population itself

Identify abnormal and coordinated behavior without waiting for labels, then connect it to real-time decisions, investigations, and strategy.

Detect without labels

Start finding risk before the first confirmed case

Identify abnormal behavior without requiring a library of known fraud examples, giving teams earlier visibility into new attacks and changing tactics.

No historical fraud labels required
Find coordinated behavior

See the pattern across the population

Analyze peer groups, behavioral dimensions, clusters, velocity, and connected activity to reveal coordinated behavior that can look normal when each event is viewed alone.

Peer groups · clusters · velocity
Adapt as behavior changes

Keep detection aligned with a moving environment

Use behavioral baselines and population-level analysis to identify meaningful deviations as legitimate and fraudulent activity evolves.

Behavioral baselines, continuously updated
Make anomalies actionable

Connect model output to the rest of the risk stack

Use unsupervised risk scores and patterns in real-time decisions, investigations, graph analysis, rules, and downstream strategies.

Scores feed decisions, cases, and rules
Product Proof

The unknown, becoming visible

Normal population behavior resolves into peer groups, a coordinated anomaly surfaces, and the result becomes risk intelligence your team can act on — without a single historical fraud label.

Behavioral
signals
→
Peer groups
& baselines
→
Anomaly
detected
→
Risk
intelligence
Behavioral signals resolve into peer groups, surface anomalies, and produce risk intelligence for real-time action.
How It Works

Learn the pattern from the data itself

1

Ingest behavioral, transaction, device, identity, and relationship features.

2

Establish peer groups, baselines, and multidimensional patterns.

3

Identify abnormal or coordinated behavior.

4

Produce risk intelligence for decisions and investigations.

5

Use confirmed outcomes to strengthen future detection where supported.

A core detection layer, not a lab experiment

Put unsupervised intelligence into live decisions

DataVisor combines patented unsupervised ML with supervised models, rules, graph intelligence, and device signals so emerging-threat detection can influence production decisions in real time.

Where labels arrive too late

Detect attacks that change faster than training data

Apply unsupervised ML to zero-day fraud, fraud rings, mule activity, account takeover, promotions abuse, payment fraud, and other attacks where historical labels provide an incomplete picture.

Zero-Day Fraud Fraud Rings & Mule Activity Account Takeover Promotions Abuse Payment Fraud
Prove the value of earlier detection

Measure what known-pattern controls miss

DataVisor's Unsupervised Machine Learning is built to catch fraud before it has a label or a rule — surfacing coordinated attacks, fraud rings, and emerging patterns that rule-based and supervised models only catch after the fact, so fraud teams get a head start that known-pattern controls alone can't provide.

Connected Platform

Give unsupervised ML more context — and more ways to act

Combine Real-Time Data Orchestration, dEdge, Knowledge Graph, Identity & Risk Signals, Rules & Features, Real-Time Decisioning, and Case Management around the unsupervised detection layer.

Real-Time Data Orchestration → Device & Behavioral Intelligence — dEdge → Knowledge Graph & Link Analysis → Identity & Risk Signals → Rules, Features & Workflow Automations → Real-Time Decisioning → Investigations & Case Management →
Explain the anomaly

Give analysts context behind the score

Every anomaly score comes with context analysts can act on — the peer group and behavioral baseline it deviates from, the transaction, identity, device, and entity signals behind it, and the cluster or network it connects to.

  • Peer-group and behavioral baseline context
  • Correlated signals: behavior, identity, transaction, device, entity
  • Cluster and network view of linked activity
  • Holistic evaluation across every signal dimension
FAQ

Common questions

Does unsupervised machine learning require labeled fraud data?

No. DataVisor's unsupervised approach is designed to identify abnormal and coordinated behavior without requiring historical fraud labels.

How does unsupervised ML reduce false positives?

By evaluating behavior in context — such as peer-group patterns, multiple signals, and relationships — unsupervised analysis can distinguish meaningful anomalies from simple one-dimensional thresholds.

Can unsupervised ML work with rules and supervised models?

Yes. DataVisor combines unsupervised intelligence with other detection layers so teams can use known-pattern controls and emerging-pattern detection together.

See an unknown attack take shape

Watch emerging fraud become detectable

Walk through how DataVisor identifies an anomalous population, connects the pattern, and turns the intelligence into action.

Request a Demo →