Unsupervised Anomaly Detection via Contingency Tables
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection solutions face issues such as excessive resource consumption, dependence on supervision, and limited universal compatibility, making them inefficient in detecting anomalies across various contexts without frequent rule maintenance and sensitivity adjustments.
Innovation Solution
A specially-configured computer system implements unsupervised universal anomaly detection by inferring its own ruleset from historical data and dynamically adjusting sensitivity based on user feedback, allowing for efficient detection across any context without significant computational power or specific configuration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural networks are used for anomaly detection, then detection accuracy is improved, but computational power consumption increases significantly
Solution Approach 1:
The system automatically infers its own ruleset from historical data without requiring external supervision or manual configuration. The anomaly detection system serves itself by learning patterns autonomously, eliminating the need for computationally intensive supervised training while maintaining detection accuracy
Solution Approach 2:
The patent replaces complex neural network mechanisms with a simpler rule-based system that uses contingency tables and restriction indexes. This substitution maintains anomaly detection capability while dramatically reducing computational power requirements
2Use of energy by moving object
If rule-based anomaly detection is used, then resource consumption is reduced, but frequent rule maintenance is required to cope with changing patterns
Solution Approach 1:
The system dynamically adapts to changing data patterns by automatically updating its inferred ruleset based on new historical data. The contingency tables and restriction indexes are continuously refined without manual intervention, allowing the system to cope with evolving patterns while maintaining low resource consumption
Solution Approach 2:
The system uses feedback from detected anomalies and user interactions to continuously improve its ruleset. By learning from false positives and negatives, the system automatically adjusts its detection criteria, eliminating the need for frequent manual rule maintenance
3Measurement precision
If supervised anomaly detection is used, then detection accuracy is improved, but universal compatibility across different contexts is limited
Solution Approach 1:
The system is designed to be universally applicable across different data types and contexts without requiring context-specific training. The contingency table approach and restriction index calculations work equally well for numerical, categorical, and mixed data types, enabling single-system deployment across diverse applications
Solution Approach 2:
The system automatically adapts to different contexts by inferring its own ruleset from the specific historical data provided for each application. This self-learning capability eliminates the need for supervised training on context-specific data while maintaining high detection accuracy across universal applications
Data Source
AI summary
Techniques for implementing unsupervised universal anomaly detection for situation handling are disclosed. In some example embodiments, a computer-implemented method comprises detecting an anomaly in a new data point that has corresponding manifestation values for variable categories based on a restriction index for the corresponding manifestation value for at least one of the variable categories in the new data point, and causing a notification of the anomaly in the new data point to be displayed on a computing device based on the detecting of the anomaly. The restriction index for the corresponding manifestation value for the at least one of the variable categories in the new data point may be calculated for the corresponding manifestation value for each other variable category in the plurality of variable categories based on a manifestation space value and a prediction space value that are based on historical data points.