Causality-Aware Root Cause Analysis for Mobile Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current root cause analysis (RCA) methods for mobile networks fail to provide actionable insights due to complexity, scale, and indirect state observability, often relying on misleading correlations and failing to account for causality relations, leading to non-deterministic and false automated responses.
Innovation Solution
A causality-aware machine learning system that determines root causes by establishing relationships between feature categories, using asymmetric SHAP to quantify impact contributions and provide actionable insights, enabling automated diagnosis and optimization of network performance without losing trustworthy aspects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ML explanation methods (naïve SHAP) are used to analyze network features, then the system can provide performance degradation analysis, but the explanations are misleading and fail to account for causality relationships between features
Solution Approach 1:
The patent introduces a causality graph as an intermediary structure that models causal relationships between network features. This graph serves as a mediator between the raw network data and the ML explanation process, enabling the system to account for causality relationships while generating explanations. The causality graph captures domain knowledge about how different network features causally influence each other, allowing the explanation system to distinguish between correlated and causally related features.
2Reliability
If hundreds or thousands of low-level network observables are monitored to detect performance degradation, then the system can identify performance issues, but the complexity of diagnosing which observables are most impactful increases significantly
Solution Approach 1:
The patent segments the large set of hundreds or thousands of low-level network observables into smaller, manageable groups using the causality graph structure. By organizing features into causal chains and groups based on their relationships in the graph, the system can analyze performance degradation in a structured manner rather than treating all features equally. This segmentation reduces diagnosis complexity by focusing attention on specific causal pathways rather than the entire feature set.
Solution Approach 2:
The patent transforms the flat, high-dimensional space of network observables into a structured causal graph with hierarchical relationships. This dimensional transformation organizes the data along causal dimensions rather than treating all features as independent variables. The causality graph introduces structural dimensions (parent-child relationships, causal chains) that reduce the effective complexity of analyzing hundreds of features by grouping them into meaningful causal structures.
3Productivity
If automated RCA systems are implemented to provide actionable insights, then the system can enable automated responses, but the responses are often non-deterministic and false due to misleading correlations
Solution Approach 1:
The patent incorporates feedback mechanisms where the causality graph structure guides the interpretation of ML explanations. The system uses the predefined causal relationships in the graph to validate and refine automated diagnoses, ensuring that identified root causes align with known causal structures. This feedback loop between the causality model and the explanation system reduces false positives and non-deterministic responses by constraining automated conclusions to causally plausible explanations.
Data Source
AI summary
A system can be configured to provide root cause analysis (“RCA”) of an issue associated with a label generated by a machine learning (“ML”) model. The system can perform operations that include determining a plurality of categories associated with a plurality of features of the ML model. The operations can further include determining a causality relationship between each category of the plurality of categories. The operations can further include determining data associated with each feature of the plurality of features. The operations can further include determining the root cause of the issue using a model explainer with ordering constraints based on the causality relationship between each category of the plurality of categories. The operations can further include performing an action associated with the issue based on the root cause of the issue.