KPI Anomaly Detection via Time Series Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The analysis of vast amounts of key performance indicator (KPI) data for communications service providers is tedious and time-consuming, requiring efficient methods for anomaly detection and visualization to extract meaningful insights.
Innovation Solution
A system for anomaly detection and analysis that processes KPI data to decompose time series data into trend, seasonality, and randomness components, using an anomaly detector to identify anomalies and expose them to web-based applications for visualization and analysis, facilitating the creation of summaries and rankings of anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of KPI data is performed using dashboards or spreadsheets, then the service provider can study data trends and seasonality, but the analysis becomes very tedious and time consuming
Solution Approach 1:
The patent replaces manual mechanical analysis methods (dashboards, spreadsheets) with an automated anomaly detection system that uses machine learning models to automatically identify anomalies in KPI data, eliminating the need for tedious manual examination while maintaining or improving analysis accuracy
Solution Approach 2:
The system enables self-service anomaly detection by automatically processing KPI data through trained machine learning models without requiring human intervention for each analysis task, allowing the service provider to continuously monitor and detect anomalies autonomously
2Loss of information
If comprehensive KPI monitoring is implemented across vast arrays of provider and user equipment, then useful insights can be obtained, but the data volume becomes vast requiring tedious analysis
Solution Approach 1:
The system extracts only the critical anomaly detection function from the vast KPI data processing requirement, using machine learning models to automatically identify and flag only the anomalous data points that require attention, rather than requiring comprehensive manual analysis of all data
Solution Approach 2:
The patent segments the complex data processing task into distinct components: data collection, anomaly detection using trained models, and result presentation. This segmentation allows each component to be optimized independently, reducing overall system complexity while maintaining comprehensive monitoring capabilities
Data Source
AI summary
An anomaly detection and analysis system detects anomalies in time series data from key performance indicators (KPIs). The system decomposes samples of the time series data received during a first time interval into a trend component, a seasonality component, and a randomness component. The system identifies an upper bound and a lower bound based on the trend component, the seasonality component, and a variance of the randomness component. The system reports a sample received after the first time interval as an anomaly when the sample exceeds the upper bound or the lower bound. The system recalculates the trend component, the seasonality component, and the randomness component when more than a threshold percentage of the samples of the time series data received during a second time interval are reported as being anomalous.