What Is Unsupervised Machine Learning?

Unsupervised machine learning (UML) is a broad category of machine learning techniques that don’t require labeled input data. Instead, UML techniques infer a function to describe the hidden structures of “unlabeled” input data points. Often used to discover patterns within large volumes of unlabeled data, UML is especially effective for discovering new and unknown patterns.

Why Is DataVisor’s Machine Learning Unique?

Detect New Attacks without Labels

UML does not require labels or training data to get started. This helps enterprises enter new markets without concern for fraud, and enables early detection and adaptive responses to fast-changing attack patterns. The UML models are self-adapting and therefore do not require constant re-tuning to maintain exceptional performance.

Handle Imperfect Data and New Values

Unlike traditional solutions that only analyze selected values from historical data, DataVisor’s UML analyzes all values, including high-cardinality categorical features and even new values from digital signals. UML is robust with respect to data quality issues, including partially missing data and data value format changes, helping enterprises unleash the full power of data.

Multi-Subspace Clustering and Graph Analysis

Clustering analysis produces suspicious groups of accounts that are highly similar or correlated. It consolidates the results by graph analysis to link clusters that share similar accounts or strong features. The process enhances the detection of unknown fraud and attack rings. 

Ensemble Modeling with Supervised Learning

The output from the unsupervised models can serve as training data to automatically train a supervised learning model and detect additional fraud. The ensemble modeling outputs a set of detected individual bad accounts, which is then combined with the detected attack rings to maximize detection coverage and reduce false positives.