ML Root Cause Analysis Using Explainable Knowledge Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Determining the root cause of performance degradation in complex systems, such as telecommunications networks, is a time-consuming and labor-intensive task that requires substantial human input, and existing machine learning (ML) techniques often fail to identify causal relationships between data features.
Innovation Solution
A method and apparatus using a trained ML model in conjunction with an ML model explainer and an ontological representation, such as a knowledge graph, to identify root causes with minimal human input by generating feature impact values and updating connections between features and prediction values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of correlated measurements from multiple domains is performed, then high-level insights and potential actions can be obtained, but the process is time-consuming and labor-intensive
Solution Approach 1:
The patent replaces manual mechanical analysis with an automated ML-based system. The system uses ML models to process measurement data, generate predictions, and automatically perform root cause analysis, substituting the manual expert analysis process with an automated computational approach that maintains high accuracy while dramatically reducing time consumption
Solution Approach 2:
The patent introduces an ML model as an intermediary between raw measurement data and root cause identification. The ML model processes the complex correlated measurements from multiple domains and translates them into actionable insights, serving as a mediator that automates the analysis process while preserving the quality of manual expert analysis
2Ease of operation
If ML techniques are used to assist operator in performing root cause analysis, then burden on operators is reduced, but existing ML techniques fail to identify causal relationships between data features
Solution Approach 1:
The patent segments the root cause analysis process into distinct functional components: data collection, ML model processing, prediction generation, and result interpretation. This segmentation allows each component to be optimized independently, with the ML model handling pattern recognition and the explanation module handling causal relationship identification, thereby reducing operator burden while preserving causal information
Solution Approach 2:
The patent introduces an ML model explainer as an intermediary component that bridges the gap between ML predictions and causal understanding. This explainer module processes the ML model's internal representations and generates human-interpretable explanations that reveal causal relationships, allowing operators to benefit from automation while still accessing causal insights
3Reliability
If end-to-end observability is implemented for Service Assurance, then service quality standards can be met, but network management complexity increases across multiple domains
Solution Approach 1:
The patent creates a universal root cause analysis system that can handle measurement data from multiple network domains (RAN, core network, transport network, etc.) through a single integrated ML model. This multi-functional approach maintains service assurance quality across all domains while reducing management complexity by providing a unified analysis platform rather than separate domain-specific tools
Solution Approach 2:
The patent merges correlated measurements from multiple network domains into a unified analysis framework. The ML model processes integrated data from various sources simultaneously, combining information from RAN, core network, transport network, and other domains into coherent root cause identifications, thereby maintaining comprehensive service assurance while simplifying the management interface
Data Source
AI summary
Methods and apparatus for root cause analysis. A method includes obtaining measurement data including measurements data of features of a system. The method further includes generating a prediction value by applying a trained machine learning model to the measurement data, and also generating feature impact values by applying a generated machine learning model explainer to the measurement data. The method further includes updating an ontological representation of connections between the features of the system and the prediction value using the generated feature impact values and, based on the updated ontological representation, outputting a proposed root cause that is responsible for the prediction value.