Anomaly Detection Using Nominal Attribute Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems are overly complex and limited in their ability to effectively detect anomalies in large data sets, particularly in insurance claim data, as they fail to adequately account for nominal attributes and lack means to improve anomaly detection beyond fraud identification in insurance claim data.
Innovation Solution
A system that utilizes a communication module to access a database of physical entity records with both numeric and nominal attributes, partitions the records based on distinct combinations of nominal attributes, and trains unsupervised anomaly detection algorithms to calculate anomaly scores, incorporating all available information for improved accuracy and seamless integration with supervised classification algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing anomaly detection systems are used, then fraud detection in insurance claim data is performed, but the systems are overly complex and fail to adequately account for nominal attributes
Solution Approach 1:
The patent segments the data processing by separating numeric attributes and nominal attributes into distinct processing streams. Numeric attributes are processed through traditional anomaly detection algorithms, while nominal attributes are processed through separate frequency analysis and pattern matching mechanisms. This segmentation allows each attribute type to be handled with appropriate methods, improving accuracy while avoiding the complexity of unified approaches that must accommodate all attribute types.
Solution Approach 2:
The patent introduces an intermediary layer that combines results from separate numeric attribute analysis and nominal attribute analysis. This intermediary integration mechanism merges anomaly scores from both streams in a controlled manner, allowing the system to leverage information from both attribute types without requiring a complex unified model that would be difficult to manage and interpret.
2Adaptability or versatility
If existing anomaly detection systems are used, then fraud detection is performed, but the systems lack means to improve anomaly detection beyond fraud identification in insurance claim data
Solution Approach 1:
The patent creates a universal anomaly detection framework that can handle multiple attribute types (numeric and nominal) and is applicable to various domains beyond insurance fraud detection. The system's architecture separates domain-specific data input from the core anomaly detection logic, allowing the same framework to be applied to different applications while maintaining reliable detection through consistent handling of mixed attribute types.
3Measurement precision
If nominal attributes are incorporated into anomaly detection, then detection precision is improved, but processing complexity increases
Solution Approach 1:
The patent segments the processing of nominal attributes into distinct frequency analysis and pattern matching components that operate independently from numeric attribute processing. This segmentation allows nominal attributes to be incorporated into anomaly detection without requiring complex integration mechanisms, as each attribute type is handled by specialized sub-systems that feed into a simple combination layer.
Solution Approach 2:
The patent employs computationally efficient methods for processing nominal attributes, such as frequency counting and simple pattern matching, rather than requiring expensive complex models. These lightweight processing mechanisms provide sufficient precision for nominal attribute analysis while keeping computational costs and processing complexity low.
Data Source
AI summary
A database includes a plurality of physical entity records of insurance claims, and each physical entity record includes physical data values for at least one numeric attribute and partition-specifying values concerning values for one or more nominal attributes from one or more insurance claim records. A method for detecting anomalies in the database includes retrieving and partitioning the plurality of physical entity records from the database, training an unsupervised anomaly detection algorithm on the plurality of physical entity records to obtain a trained anomaly detection model for each partition, calculating an anomaly score for each physical entity record using the trained anomaly detection model associated with each partition, and updating the plurality of physical entity records in the database by adding the associated anomaly score. The method is used to determine if a user-provided physical entity record is fraudulent using the anomaly score.