Context-Sensitive Anomaly Rules for Network Outlier Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in communications networks fail to provide context-sensitive and generalizable rules for filtering anomalous values, leading to incorrect explanations and inadequate management of network performance due to concept drift and lack of use case-specific feedback.
Innovation Solution
A method involving a first node that generates context-sensitive rules for detected outliers by comparing their values to reference statistics within specific contexts, and a second node that applies these rules to newly detected outliers, enabling efficient and robust anomaly handling through active rule mining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing anomaly detection methods are used, then anomalies can be detected, but the rules generated are not context-sensitive and lack generalizability, leading to incorrect explanations
Solution Approach 1:
The patent applies local quality by creating context-specific reference statistics for different network conditions. Instead of using uniform anomaly detection thresholds across all scenarios, the system generates context-aware reference statistics that adapt to local network conditions, time periods, and traffic patterns, thereby preserving context information while maintaining detection accuracy
Solution Approach 2:
The patent implements dynamics by making the anomaly detection rules adaptive and evolving. The system continuously learns from new data, updates reference statistics dynamically, and refines rules based on feedback. This dynamic approach allows the detection system to adapt to changing network conditions while maintaining generalizability across different contexts
2Reliability
If existing anomaly detection methods are used, then anomalies can be detected, but the system cannot adapt to concept drift and changing network conditions
Solution Approach 1:
The patent applies feedback by implementing a closed-loop system where detection results, operator confirmations, and performance metrics are fed back into the rule refinement process. This feedback mechanism allows the system to continuously improve its rules, adapt to concept drift, and maintain reliability while becoming more versatile in handling changing network conditions
Solution Approach 2:
The patent implements preliminary action by proactively updating reference statistics and refining rules before significant concept drift occurs. The system performs preliminary learning and adaptation phases that prepare the anomaly detection mechanism for upcoming changes in network behavior, thereby maintaining reliability during transitions
3Measurement precision
If context-sensitive rules are generated for each outlier, then anomaly filtering becomes more accurate, but the complexity of rule generation and management increases
Solution Approach 1:
The patent applies universality by creating a multi-functional rule generation system that handles multiple anomaly types, contexts, and network conditions through a unified framework. The system uses universal statistical methods and learning algorithms that can adapt to different scenarios, thereby maintaining high filtering precision without proportionally increasing management complexity
4Measurement precision
If multivariate AD algorithms are used, then complicated anomalies can be modeled, but the system cannot directly explain why they occur
Solution Approach 1:
The patent applies segmentation by breaking down complex multivariate anomalies into component parts and analyzing them through context-specific reference statistics. The system segments the anomaly detection process into rule generation, rule application, and explanation phases, allowing it to maintain sophisticated detection capabilities while providing interpretable explanations for each detected anomaly
Data Source
AI summary
A method, performed by a first node (111), for handling anomalous values. The first node (111) generates (305) a respective rule for each outlier in a set of outliers. Each respective rule comprises a respective set of conditions. Each condition compares a respective value of a respective variable to a corresponding respective reference statistic for a respective context corresponding to the respective value. The first node (111) determines (306), for each generated respective rule, whether or not it matches a previously unappraised rule. The first node (111) also initiates (307) performing one of: i) with the proviso that the generated respective rule is determined to match a previously unappraised rule, adding one count, and ii) providing, with the proviso that the generated respective rule is determined to lack a match to a previously unappraised rule and to any previously appraised rule, an indication. The indication indicates the generated respective rule is unappraised.