Machine learning outlier detection using weighted histogram-based outlier scoring (w-hbos)

An automatic outlier detection and analysis system, comprising:

• a processor;
• and
• a computer-readable medium on which is stored machine-readable instructions that cause the processor to:
• access a selected set of features extracted from merchant data where the merchant data includes entities and transactions conducted by the entities;
• obtain a higher value outlier subset from a set of outliers generated by a trained outlier identification model using a transformed, selected set of features obtained from the merchant data, where the trained outlier identification model implements a Weighted Histogram-based Outlier Scoring (W-HBOS) model;
• determine as a top reason code for the higher value outlier subset, a feature from the selected set of features;
• and
• output entities determined from the merchant data that are ranked in accordance with corresponding contributions to the feature selected as the top final reason code.

A non-transitory processor-readable storage medium comprising machine-readable instructions that cause a processor to:

• identify a selected set of features by filtering an initial set of features extracted from real-world data via an intermediate feature selection;
• generate a transformed set of features by transforming the selected set of features via orthogonal feature transformation;
• select a subset of outliers from at least two sets of outliers obtained from the real-world data by providing the transformed set of features to a trained outlier identification model, where the trained outlier identification model is implemented as a Weighted Histogram-based Outlier Scoring (W-HBOS) model;
• identify one or more features from the initial set of features with higher contributions to outliers in the selected subset of outliers;
• and
• output one or more entities ranked in accordance with contributions to the one or more features.