20240028616. EXPLAINABLE LAYERED CONTEXTUAL COLLECTIVE OUTLIER IDENTIFICATION IN A HETEROGENEOUS SYSTEM simplified abstract (Kyndryl, Inc.)

EXPLAINABLE LAYERED CONTEXTUAL COLLECTIVE OUTLIER IDENTIFICATION IN A HETEROGENEOUS SYSTEM

Organization Name

Kyndryl, Inc.

Inventor(s)

Mouleswara Reddy Chintakunta of Allagadda (IN)

EXPLAINABLE LAYERED CONTEXTUAL COLLECTIVE OUTLIER IDENTIFICATION IN A HETEROGENEOUS SYSTEM - A simplified explanation of the abstract

This abstract first appeared for US patent application 20240028616 titled 'EXPLAINABLE LAYERED CONTEXTUAL COLLECTIVE OUTLIER IDENTIFICATION IN A HETEROGENEOUS SYSTEM

Simplified Explanation

The present invention provides a method for identifying outliers in a heterogeneous system by detecting and explaining the outliers. The method involves receiving input data from various data sources with different data types and converting them into a standardized format. Global outliers are detected in the first pass of the data, followed by the detection of contextual outliers in the second pass. The global and contextual outliers are then grouped based on their type. Finally, output data is generated, including explanations for each detected outlier.

• The invention detects outliers in a heterogeneous system.
• It converts input data from different sources and types into a standardized format.
• Global outliers are identified in the first pass, and contextual outliers are identified in the second pass.
• The detected outliers are grouped based on their type.
• The output data includes explanations for each detected outlier.

Potential Applications

• Anomaly detection in various industries such as finance, healthcare, and cybersecurity.
• Quality control in manufacturing processes.
• Fraud detection in financial transactions.
• Identifying abnormal behavior in network traffic.

Problems Solved

• Identifying outliers in a heterogeneous system with diverse data sources and types.
• Providing explainability for detected outliers.
• Efficiently detecting both global and contextual outliers.

Benefits

• Improved accuracy in outlier detection.
• Enhanced understanding of the reasons behind outliers.
• Streamlined data processing and analysis.
• Early identification of anomalies for timely intervention.

Original Abstract Submitted

embodiments of the present invention provide an approach for identifying outliers (e.g., detecting the outliers and generating outlier explainability) in a heterogeneous system. heterogeneous input data is received from any number of data sources having any number of data types and converted into a single predefined format. global outliers are detected in a first pass of the data. contextual outliers are detected in a second pass. global and contextual outliers are then collectively grouped based on outlier type. output data is then generated including explainability for each detected outlier.