KPI Root Cause Identification Using Spectral Saliency Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing root cause analysis methods for complex systems, particularly in telecommunication core networks, are time-consuming, error-prone, and unsuitable for large-scale KPI analytics, making it difficult to identify the root causes of anomalous incidents efficiently.
Innovation Solution
A method and apparatus that utilize spectral saliency analysis by calculating the phase spectrum of KPI time sequences using discrete Fourier Transform, applying linear transformations to prevent spectral leakage, and employing time-lagged cross-correlation to identify secondary KPIs with the greatest similarity to primary KPIs as root causes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional root cause analysis methods are used in complex systems, then comprehensive analysis of all KPIs can be performed, but the process becomes extremely time-consuming and difficult
Solution Approach 1:
The patent extracts only the most relevant secondary KPIs that have causal relationships with the primary KPI anomaly, rather than analyzing all KPIs in the system. This is achieved by calculating spectral saliency values and selecting top-N secondary KPIs with the highest values, thereby extracting the critical subset of data needed for root cause identification while discarding irrelevant information.
Solution Approach 2:
The patent introduces spectral saliency analysis as an intermediary mechanism between primary KPI anomalies and secondary KPI selection. The spectral saliency values act as a mediator that quantifies the causal relationship strength, enabling automated selection of relevant secondary KPIs without manual intervention or expert knowledge.
2Measurement precision
If manual root cause analysis is performed by experts, then accurate identification can be achieved, but sophisticated expertise is required and the process remains time-consuming
Solution Approach 1:
The system performs self-service by automatically identifying root causes through automated spectral saliency calculation and secondary KPI selection. The automated processing eliminates the need for expert intervention while maintaining high accuracy in root cause identification, allowing the system to serve itself without sophisticated human expertise.
Solution Approach 2:
The patent replaces the mechanical process of expert analysis with an automated computational system based on spectral analysis. The mechanical expertise required for manual analysis is substituted by an automated algorithm that calculates spectral saliency values and identifies relevant secondary KPIs through mathematical processing rather than human judgment.
3Measurement precision
If all secondary KPIs are analyzed to identify root causes, then comprehensive coverage is achieved, but computational overhead increases significantly
Solution Approach 1:
The patent applies partial action by analyzing only the top-N secondary KPIs with the highest spectral saliency values rather than all secondary KPIs. This selective approach achieves sufficient root cause identification accuracy while significantly reducing computational overhead by processing a subset of the most relevant data points.
Solution Approach 2:
The patent changes the parameter of analysis scope from comprehensive (all secondary KPIs) to selective (top-N based on spectral saliency). By transforming the parameter of data volume and using spectral saliency as a filtering criterion, the system reduces computational requirements while maintaining effective root cause identification.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A method for identifying one or more root causes of an anomalous incident associated with a primary Key Performance Indicator (KPI) in a system is disclosed. The method includes selecting a plurality of time sequences of secondary KPIs of the system by a processor (104A-N), for calculating a spectral saliency of a time sequence of the primary KPI and each secondary KPI time sequence. The method includes calculating similarities between the time sequence of the primary KPI and each secondary KPI time sequence based on the spectral saliencies for selecting one or more secondary KPIs having greatest similarity to the primary KPI as root cause KPIs by the processor (104A-N). The one or more root causes are identified quickly, accurately, and efficiently by correlating the anomalous primary KPI with the most relevant secondary KPIs regardless of their time series characteristics.