Anomaly Root Cause Analysis Using Dual Bounding Boxes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods in wireless networks struggle to accurately identify the root cause of anomalies, particularly in multi-dimensional data sets, where subtle changes in variable relationships and extreme magnitudes can lead to unexplained anomalies, making it difficult to correct or address network issues effectively.
Innovation Solution
The method involves constructing both a magnitude bounding box and a principal component analysis (PCA) bounding box to classify anomaly data points, determining their type (individual magnitude, relationship, or joint magnitude anomalies) based on their position relative to these boxes, and using PCA to identify the specific variables involved in the anomaly, thereby facilitating root cause analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional anomaly detection methods are used, then anomalies can be detected, but the root cause cannot be accurately identified
Solution Approach 1:
The patent segments the anomaly analysis process into two distinct bounding box evaluations: magnitude bounding box for individual variable extremes and PCA bounding box for relationship anomalies. This segmentation allows precise identification of anomaly types and their root causes by comparing the anomaly point against both bounding boxes independently.
Solution Approach 2:
The patent introduces PCA (Principal Component_analysis) as an additional dimensional perspective beyond simple magnitude thresholds. By transforming data into principal component space, the system can detect anomalies in variable relationships that are not apparent in the original feature space, thus identifying root causes that traditional methods miss.
2Measurement precision
If multiple bounding boxes are constructed for classification, then anomaly classification precision is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing the two bounding boxes (magnitude and PCA) during a training phase. During anomaly detection, the system only needs to compare the anomaly point against these pre-computed bounding boxes, avoiding complex real-time calculations and reducing operational computational complexity.
Solution Approach 2:
The system uses the data's own statistical properties (mean, standard deviation for magnitude bounding box; principal components for PCA bounding box) to automatically define the bounding boxes without requiring external manual configuration or complex algorithms, enabling the system to self-configure for different data sets.
Data Source
AI summary
A method includes receiving an anomaly data point and comparing the anomaly data point to a magnitude bounding box to produce a first comparison. The method also includes comparing the anomaly data point to a principal component analysis (PCA) bounding box to produce a second comparison and classifying the anomaly data point in accordance with the first comparison and the second comparison to produce a classification.