Classifier Generation Apparatus for Abnormal Data Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for identifying patterns, such as those using Support Vector Machines (SVM) and 1-class SVM, face challenges in accurately distinguishing between specific and non-specific category patterns, especially when there is a small number of non-specific category patterns used for learning, leading to potential misclassification of abnormal situations as normal.
Innovation Solution
A classifier generation apparatus that calculates a specific category likelihood for input data by using a predetermined parameter, acquired through feature extraction from both specific and non-specific category data, to determine whether the input data belongs to the specific category, with a focus on decreasing the likelihood of non-specific category data being misclassified as specific category data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a 2-class classifier (SVM) is used to separate specific and non-specific category patterns, then the classifier can identify both categories, but when there is a small number of non-specific category patterns used for learning, the classifier may erroneously determine non-specific category patterns as specific category patterns
Solution Approach 1:
The invention segments the learning process into two distinct parts: first training a 2-class classifier to separate specific and non-specific categories, then training a separate 1-class classifier to model the distribution of specific category patterns. This segmentation allows each classifier to specialize in its respective function, improving overall reliability even with limited non-specific category data.
Solution Approach 2:
The invention merges the outputs of two classifiers: the 2-class classifier that distinguishes between specific and non-specific categories, and the 1-class classifier that models the specific category distribution. By combining these classifiers, the system achieves both the ability to separate categories and the ability to handle limited non-specific category patterns effectively.
2Reliability
If a 1-class classifier is used to model specific category patterns, then the classifier can handle cases with limited non-specific category data, but non-specific category patterns cannot be used for learning, making it impossible to heighten detection sensitivity to known non-specific category patterns
Solution Approach 1:
The invention segments the learning objectives into two parts: the 1-class classifier focuses on modeling the specific category distribution for stability, while the 2-class classifier focuses on learning to distinguish non-specific patterns for sensitivity. This segmentation allows each classifier to optimize for its specific purpose without compromise.
Solution Approach 2:
The invention merges the strengths of both classifier types by combining their outputs. The 1-class classifier provides stable modeling of specific categories, while the 2-class classifier contributes enhanced detection sensitivity to non-specific patterns, achieving both reliability and precision simultaneously.
3Measurement precision
If asymmetric SVM is used to maximize margins between specific and non-specific patterns, then detection sensitivity to known patterns is improved, but unknown non-specific category patterns are likely to be determined as specific category patterns
Solution Approach 1:
The invention segments the pattern recognition task into two specialized classifiers: the 2-class classifier (asymmetric SVM) handles known pattern discrimination with high sensitivity, while the 1-class classifier provides a general model of specific category patterns that helps identify unknown non-specific patterns. This segmentation resolves the contradiction between sensitivity to known patterns and adaptability to unknown patterns.
Solution Approach 2:
The invention merges the asymmetric SVM's ability to maximize margins for known patterns with the 1-class classifier's ability to model the overall specific category distribution. This combination maintains high detection sensitivity to known non-specific patterns while improving adaptability to identify unknown non-specific category patterns that deviate from the specific category distribution.
Data Source
AI summary
There is provided a classifier capable of classifying unknown abnormal data input to the classifier even if there is a small number of abnormal data used for the learning of the classifier. When learning parameters of the classifier, the specific category likelihood of normal patterns for learning relatively deviating from a group of normal patterns for learning is decreased relatively to the specific category likelihood of normal patterns for learning not relatively deviating from the group of normal patterns for learning, and the specific category likelihood of abnormal patterns for learning is decreased relatively to the specific category likelihood of the group of normal patterns for learning.