Anomaly Detection Feature Selection Bias Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional anomaly detection methods face accuracy declines due to bias in training data and local bias in feature space, leading to reduced detection accuracy.
Innovation Solution
An anomaly detection device that selects features based on attached information rather than actual training data, reducing bias and enhancing accuracy by calculating anomaly degree data using selected first-type features and second-type features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If plural similar pieces of training data are used for anomaly detection, then the detection process can be performed, but bias in the training data causes a decline in detection accuracy
Solution Approach 1:
The patent extracts and removes biased training data samples from the dataset. The selection unit identifies and extracts samples with high bias based on bias evaluation results, then excludes them from the anomaly detection process. This extraction of harmful elements directly resolves the contradiction by eliminating the source of accuracy degradation while preserving the detection capability using the remaining unbiased samples.
Solution Approach 2:
The patent changes the parameter of training data composition by dynamically adjusting which samples are included in the training set. Based on bias evaluation metrics, the system modifies the training dataset parameters to exclude biased samples, thereby transforming the quality characteristics of the training data to achieve both detection capability and high accuracy.
2Productivity
If feature dimension reduction is performed on training data, then processing efficiency is improved, but local bias in the training data is not held down, resulting in a decline in detection accuracy
Solution Approach 1:
The patent performs preliminary bias evaluation and sample extraction before the anomaly detection process. By pre-identifying and removing biased samples from the training data, the system prepares a high-quality dataset in advance. This preliminary action ensures that subsequent dimension reduction and detection processes work with unbiased data, maintaining both efficiency and accuracy without needing to reprocess biased samples.
Data Source
AI summary
According to one embodiment, an anomaly detection device includes a feature calculating unit, a first selecting unit, an anomaly degree data calculating unit. The feature calculating unit calculates or refers to a first-type feature of each of plural pieces of training data, and calculates or refers to a second-type feature of target data for detection. The first selecting unit selects, based on first-type attached information corresponding to each of the plural pieces of training data, at least one or more of plural first-type features. The anomaly degree data calculating unit calculates anomaly degree data indicating a degree of anomaly in the target data for detection, using the selected first-type feature and using the second-type feature.